Novel Rehabilitation Rocking Chair for Trunk Muscle Activation and Pattern Analysis in Children with Spinal Cord Injury.

NOVEL REHABILITATION ROCKING CHAIR FOR TRUNK MUSCLE ACTIVATION

AND PATTERN ANALYSIS IN CHILDREN WITH SPINAL CORD INJURY

CHAPTER 1

INTRODUCTION

As in adults, spinal cord injuries (SCI) for pediatric patients are often

catastrophic, leading to dramatic changes in quality of life, overall health, and life

expectancy, not to mention the emotional and financial burdens carried by the families of

these patients. However, as younger SCI patients grow, additional medical complications

arise that are unique to this population due to musculoskeletal immaturity, i.e. the

patients are still growing. Conditions such as such as scoliosis and pneumonia can be

generally attributed to poor trunk control, a term used to describe the ability for a patient

to sit upright against the pull of gravity. Paradigm-shifting therapies such as locomotor

therapy (LT) have recently been developed as a mechanism to engage these patients in an

activity that enhances the quality of life for both adults and children living with SCI.

Therapies that improve trunk control, for pediatric SCI patients, is a critical therapy

target. As this dissertation will explore, LT has been shown to be a reliable method of

improving trunk control in a clinical setting. Extending the ability to make gains beyond

the clinic through the development of technology appropriate in the home environment is

the primary focus of this dissertation research project. In particular, this dissertation

outlines the development, evaluation, and analysis of a sensorized rocking chair designed

specifically for children with spinal cord injuries.

1.1 Project Goals

The goals of this multifaceted research project included: 1) the development of a

multi-sensor equipped prototype rocking chair for use by pediatric participants age 1-12

years with neurologic deficits that impair upright trunk control, 2) the characterization of

trunk muscle activation when children with SCI rock, 3) comparison of muscle activation

in TD children to muscle activation seen in children with impaired trunk control, 4)

characterization the output of sensors integrated in the rocking chair when children with

different levels of trunk control rock, and 5) the analysis of data produced by the rocking

chair sensors to find correlations between muscle group usage patterns, and to produce an

evaluation mechanism that clinicians can use to evaluate patient progress.

1.2 Project Specific Aims

To create a safe and engaging rocking chair for the pediatric patient population

that incorporates mechanisms to investigate and assess trunk activity levels in a manner

that is helpful to clinicians, this study has been broken down into four specific aims.

Aim 1: Design and fabricate an instrumented rocking chair for children with impaired

trunk control due to spinal cord injury and verify that it meets safety and operational

criteria.

This aim focuses on fabricating a rocking chair to provide therapists with a safe,

enjoyable method to enable trunk movement in children with impaired trunk control for

integration into activity-based therapy, or for use in increasing activity levels in the home.

Additionally, it will enable research to discover patterns of muscle activation during

rocking and can be evaluated for its ability to provide data which can be used in

assessment of trunk control in this population.

Aim 2: Characterize muscle activation in typically developing children and in those with

impaired trunk control while rocking in a rocking chair and describe differences

between muscle activation in the two groups.

This aim focuses on investigation of muscle activation during rocking in children

with SCI to determine if rocking activates muscles of interest, and to characterize muscle

activation patterns. This will help to investigate the hypothesis that rocking activates

trunk muscles, and to provide information about muscle activation timing in typically

developing children and children with SCI.

Aim 3: Characterize the relationship between rocking chair dynamics and Segmental

Assessment of Trunk Control (SATCo) score when children with different trunk control

capabilities rock.

This aim focuses on defining how the output of sensors changes when children

with different levels of trunk control perform the rocking activity, and identifying the

muscles in different domains (legs, arms, trunk) used during the activity. These will offer

insight into the mechanisms of how a child initiates and maintains the rocking activity

and will provide information which can be used to investigate other methods of

assessing trunk control.

Aim 4: Develop an algorithm to assess trunk control using data produced by sensors

on the instrumented rocking chair and evaluate its clinical utility.

This aim focuses on developing methods to utilize data collected from rocking

chair sensors to provide therapists with useful information about the child’s trunk control.

Regression analysis will used to find correlations between sensor data and trunk control,

and to produce a model for predicting SATCo score. If, however, after checking

correlations, there are not any independent variables which are well correlated with

SATCo other methods of data processing may be explored.

Successful completion of these aims sets the stage for future research to

investigate the clinical use of the chair both as a diagnostic tool (perhaps initially as a

confirmation of the existing SATCO assessment technique in the clinic), a tracking tool

(to monitor patient gains made through LT applied in the clinic), and possibly as a means

of applying activity-based therapy in the home environment.

CHAPTER 2

BACKGROUND

2.1 Demographics of Spinal Cord Injury

The National Spinal Cord Injury (SCI) Statistical Center, estimates that there are

currently 299,000 people living with SCI in the United States, with about 18,000 new

cases of SCI occurring annually [1, 2]. Of these, approximately 4-5% of are children at

the time of injury [3]. The most prevalent cause of pediatric SCI is motor vehicle

accidents (32%), followed by falls (18%) [4].

The economic impact of SCI can be devastating. SCI persons injured since 1970

spent an average of 171 days in a hospital over the first 2 years post injury [5]. In 2001,

Sekhorn et al reported that initial hospital expenses averaged $95,203, and $2958 per

year after initial recovery and rehabilitation. Other medical services and equipment

averaged $4,908 per year, and personal assistance costs and costs of institutional care

averaged $6,269 per year. In addition, loss of productivity and income are major factors

for consideration especially in the case of young people and children [6].

2.2 Spinal Cord Physiology

The spinal cord is the primary information pathway that receives sensory

information from the body and relays it to the brain. It also carries messages from the

brain to other body systems [7]. In addition to transmitting information, the spinal cord

actively processes information and generates motor outputs. Millions of nerve cells

collected into neuronal networks, referred to as the central pattern generator, are situated

in the spinal cord and coordinate complex movement patterns such as rhythmic breathing

and walking [8].

The spinal cord is linked to the muscles via spinal nerves which exit the spine at

different levels and nerve function is determined by the level as shown in Figure 1. The

cervical spinal nerves (C1 to C8) control the back of the head, the neck and shoulders, the

arms and hands, and the diaphragm; the thoracic spinal nerves (T1 to T12) control the

chest muscles, some muscles of the back, and many organ systems, including parts of the

abdomen; the lumbar spinal nerves (L1 to L5) control the lower parts of the abdomen and

Figure 1: Spinal nerve origins and functions

Adapted from Henry Gray, Anatomy: descriptive and surgical (1858)

the back, the buttocks, some parts of the external genital organs, and parts of the leg;

finally, the sacral spinal nerves (S1 to S5) control the thighs and lower parts of the legs,

the feet, most of the external genital organs, and the area around the anus [7].

2.3 Spinal Cord Injury

The most common and severe cause of spinal cord injury (SCI) is trauma.

Traumatic injuries (falls, etc.) often fracture vertebral bodies which can lead to

compression of the spinal cord, and associated permanent impairments including motor,

sensory and autonomic dysfunction. Other causes of SCI include inflammatory

conditions, infections, vascular issues, and neoplastic or degenerative changes [9]. The

nature and severity of dysfunction depends largely on the level of the spine at which the

injury occurs, and the degree to which the spinal cord is damaged [9, 10].

Failure to recover post injury leads to neurological disability and is the result of

axonal and other cellular damage, followed by regenerative failure and loss of

neuroplasticity, the ability of the nervous system to alter activity in response to stimuli by

reworking structure, functionality, and/or connections, in the region of the injury [9].

Recovery of function after injury depends upon the severity of neurological injury, with

cell and axon survival leading to proportionately improved recovery, and also the

neurological level of injury, with higher anatomical injuries leading to greater loss of

function [7, 10]. It has typically been believed that most neurological recovery in patients

with SCI occurs within the first 6 months after injury, but it has been established that

improvements can continue to occur years later [10].

2.4 Impacts of Spinal Cord Injury

The long-term effects of SCI are far reaching and affect almost every body

system, including respiration, head/trunk control, bladder control, speech, cardiovascular

function, and the musculoskeletal system [11]. Table 1 includes an abbreviated list of

conditions either caused by or aggravated by SCI and the prevalence in a cohort of

patients whose initial injury occurred during childhood [12].

Table 1: Common complications associated with SCI

Complications

Number with

complication

Number of

Respondents

Percent of

Respondents

Urinary tract infections 160 215 74%

Severe UTI 41 214 19%

Urinary stones 54 215 25%

Bladder incontinence 52 212 25%

Orchitis/ epididymitis 15 147 10%

Autonomic dysreflexia 85 157 54%

Hyperhidrosis 31 210 15%

Bowel incontinence 135 215 63%

Latex allergy 18 208 9%

Pressure ulcers 94 216 44%

Thromboembolism 41 216 19%

Respiratory complications 71 213 33%

Chronic medical conditions 42 214 20%

Other hospitalizations 59 215 27%

Adapted from Vogel et al 2002

2.5 Functional Impairments

In a survey of 681 adults with spinal cord injuries, 7 functions were ranked with

respect to which would most dramatically improve the respondent’s life if it was

regained. The functions included arm/hand function, upper body/trunk strength,

bladder/bowel function, sexual function, elimination of chronic pain, normal sensation,

and walking movement. Results showed that those with paraplegia ranked sexual

function and bladder/bowel function as the first and second highest priority, and those

with tetraplegia ranked arm/hand function and sexual function as the first and second

highest priorities. Regaining trunk stability was rated the third highest of the seven

functions most desired by both groups to improve their quality of life [13].

2.6 Spinal Cord Injury Rehabilitation

Physical rehabilitation, which is the primary intervention post-SCI for both adults

and children, has traditionally focused on management and compensation as the primary

solutions to combat impaired trunk control [14, 15]. Compensation in this context refers

to the use of assistive technology devices often specifically designed for SCI patients,

and management refers to medical interventions to treat medical conditions associated

with SCI.

Traditional approaches have involved the use of support mechanisms such as

thoracolumbosacral orthoses [16, 17], supportive and adaptive seating systems, cushions,

and chest straps [14, 18]. Additionally, practical techniques such as hooking an arm over

a wheelchair handle or positioning arms to create a stable base of support can help to

compensate for lack of trunk control [14]. These compensation-based strategies can help

the patient to engage in functional activities, but do not focus on neuromuscular

function. Moreover, by restricting movement they may contribute to deconditioning and

reduction in muscle mass, eventually leading to further impairment as neural circuits in

the spinal cord become inactive with disuse [19, 20].

Task specific, activity-based training methods have been investigated in adults

with thoracic SCI, to improve intrinsic trunk control by actively engaging the

neuromuscular system through sensorimotor input. Methods that have been studied

include kayak ergometer training [21], exercises that move the upper body outside the

base of support [22], and balance exercises performed on a rocker board which required

subjects to engage their trunk muscles to maintain balance and stability [23]. These

studies have demonstrated improvement in various functional measures of balance and

trunk control, and provide evidence that it is possible to improve intrinsic trunk control in

adults with SCI.

Recently, however, a new paradigm has emerged which promises the possibility

of restoration of at least some intrinsic neuromuscular control. In the context of a healthy

spinal cord, an input signal, which can be interpreted as the intent to perform a physical

action, is provided by the brain, and arrives in the central pattern generator of the spinal

cord via descending neural pathways. In the context of a compromised spinal cord, where

the descending pathways have been interrupted (corrupted, damaged, completely

severed), it has been shown that sensory input ascending to the spinal cord can instead (or

in place of the descending signal) stimulate motor responses, and that the spinal cord can

actually be retrained to respond by generating appropriate motor responses [24].

2.7 Pediatric Spinal Cord Injury

2.7.1 Trunk Control

Children with SCI experience many of the same consequences as those seen in

adults, but in addition to the impact of the initial injury, there are also debilitating

secondary consequences including altered growth and development as they mature [25].

In children as in adults, the long-term effects of SCI are far reaching and affect almost

every body system, including respiration, head/trunk control, bladder and bowel control,

speech, cardiovascular function, and the musculoskeletal system [11, 26, 27].

Skeletal immaturity and continuous musculoskeletal growth in the context of

impaired trunk control can also place children with SCI at substantial risk for developing

chronic conditions such as pneumonia, and scoliosis, a likelihood that increases the

younger the age of injury [3, 13, 17, 27, 28]. In a retrospective study by Dearolf et al, of

130 children with spinal cord injuries sustained before the typical adolescent growth spurt

(defined as age <12 years for girls and <14 years for boys), 97% developed scoliosis of

greater than 10 degrees. By way of comparison, only 48% of those injured after the

growth spurt developed scoliosis [17].

Surgical intervention to correct scoliosis (Figure 2) in particular, is a highly

invasive procedure that increased in cost from approximately $72,000 in 2001 to

$155,000 in 2011, with the cost of spinal implant mechanisms suggested as the primary

reason for the annual 11.3% increase [29]. Further, surgery to correct for scoliosis is often

Figure 2: Representative image of scoliosis, pre (left) and post-surgery (right).

Source:

pediatricscoliosissurgery.com/case-studies/meghan-pediatric-scoliosis-case/

repeated as a child ages, can result in further restriction of movement, and often interferes

with development of lung capacity in younger children [30].

2.7.2 Compromised Respiratory Function

Impairment of respiratory muscles in pediatric SCI can lead to respiratory

insufficiency [17, 28, 31, 32], and impairment of trunk muscles leads to the development

of neuromuscular scoliosis in most children who sustain SCI prior to skeletal maturity.

This can decrease the mechanical efficiency of the chest wall, further reducing lung

capacity and function [3, 28, 32]. Furthermore, due to years of immobility and lack of

weight bearing, pathophysiological changes in the musculoskeletal system progress, and

respiratory function is further compromised [27, 31, 32]. Respiratory insufficiency can

ultimately lead to complications such as infection, pneumonia (Figure 3) and even death

[32].

Figure 3: Image of bacterial pneumonia (right lobe of the lung)

Source: emedicine.medscape.com/article/967822-overview

2.7.3 Trunk Control Assessment

The Segmental Assessment of Trunk Control (SATCo) scoring tool was

specifically developed for use in pediatric populations to evaluate the trunk incrementally

across seven levels from shoulder support to no support and for three types of control

(static, active and reactive) [33, 34]. With each incremental level of external support,

trunk control above the location of support is evaluated across the control types and if the

child is deemed competent, the assessment continues until the child is unable to maintain

an appropriate sitting posture. A SATCo score of ‘0’ indicates no head control, whereas a

perfect score of ‘20’ indicates that a child can sit without any support and control their

head and trunk during static, active, and dynamic tests of control. Figure 4 shows a

SATCO assessment form designed for clinical use [35].

2.8 Rehabilitation

Various assistive technologies and therapeutic interventions have been used to

address impaired trunk control for children with SCI. During rehab sessions, patients

learn specific methods to compensate for their impairment. For instance, to balance and

sit upright by using counter-balance maneuvers with the arms and head above their

passive trunk and to establish an adequately wide base of support (i.e. posterior pelvic tilt

base). In some cases, the torso may be propped to a weight-bearing position using

extended arms. practical techniques such as hooking an arm over a wheelchair handle or

positioning arms to create a stable base of support can help to compensate for lack of

trunk control [10]. These compensation-based strategies can help the patient to engage in

functional activities, but do not focus on neuromuscular function.

Figure 4: SATCo testing procedure.

Rehabilitation interventions for children with SCI may also utilize devices such as

standers, wheelchairs with supportive and adaptive seating systems, cushions, and chest

straps [14, 18], and braces such as thoraco-lumbo-sacral orthoses (Figure 5) [16, 17].

While these devices provide support, many of them can also restrict trunk, leg, and body

movement. Medications to manage spasticity, such as botox injections and baclofen can

also add to paralysis, and further prevent movement [36]. By restricting movement some

strategies may contribute to deconditioning and reduction in muscle mass, eventually

Figure 5: Thoraco-lumbo-sacral orthoses

leading to further impairment as neural circuits in the spinal cord become inactive with

disuse [19, 20, 33].

From historical evidence and general clinical observation, pediatric rehabilitation

professionals do not expect or even suggest that therapeutic interventions promote or

restore trunk control in children post-SCI [33], and after 12 to 18 months post injury,

additional recovery is generally not expected, and parents are often given little to no hope

for recovery [36].

2.8.1 ABT for Children with SCI

Based on spinal cord neuroplasticity, a new field of therapy for SCI has emerged,

which is referred to as Activity-Based Therapy (ABT). Currently, in the pediatric

recovery-based program at the University of Louisville, the primary therapeutic

intervention for pediatric SCI subjects is activity-based locomotor training (AB-LT)

conducted 1.5 hours per day, five days per week with a targeted goal of 60 sessions over

roughly three months of therapy. Therapy sessions consist of a minimum of 55 minutes

on a walking treadmill equipped with active body weight support (BWS), where the

patient is assisted by four trained physical therapists, where the treadmill and body

weight support are integrated into the unit (Figure 6), providing therapeutic intervention

for SCI patients. The intervention delivered through use of the treadmill activity

provides task-specific sensory input that reinforces upright, appropriate alignment of the

trunk,

Figure 6: Pediatric locomotor training system.

Courtesy of Frazier Rehab and PowerNeuro Recovery.

pelvis and legs, weight bearing through legs, age-appropriate walking speeds, arm swing

and inter/intra-limb coordination of kinematics for both standing and stepping [36].

AB-LT therapy is followed immediately by 30 minutes of assessment by the

therapists and clinicians as well as off-treadmill activities that further emphasize and

reinforce the gains attained/observed from repeated application of the specialized

training. These extended activities promote active trunk extension or rotation by placing

balls or toys overhead or to the child’s side for them to reach for and grasp. Postures that

minimize use of compensation are also encouraged by having children shoot baskets or

carefully remove pieces of a puzzle tower while maintaining a stable trunk. Encouraging

children to use postures such as arms in contact with the body or arms extended behind

the body can also help to minimize compensation [37, 38].

2.8.2 Trunk Control Improvements with ABT

In children with SCI, research has shown that activity-based locomotor training

(AB-LT), which provides sensorimotor input to activate the neuromuscular system,

leads to improvements in intrinsic trunk control as measured by SATCo score.

Significant changes in SATCo scores (p <0.0001) were determined for all participants

from initial to post-60th session evaluation [33, 34]. This improvement is observed

regardless of the chronicity or severity of the initial impairment, with similar

improvements seen in cervical, and in high and low thoracic injury [33]. These findings

align with case studies that support the notion of sensorimotor input playing a crucial

role in enhancing trunk control in children with SCI [39-41].

2.9 Rocking Chair to Activate Trunk Muscles in Pediatric SCI

Studies demonstrate that improving intrinsic trunk control in subjects with SCI is

possible [21-23, 33, 39-41]. There is need, however, for methods to reinforce and

maintain these gains in trunk control through community integration activities that allow

patients to continue to extend their capacity via the retrained nervous system in the home

and community [21].

One activity that researchers have explored is the act of rocking in a rocking

chair, which enables a user to generate movement with minimal or no caregiver

assistance. In one case study, a pediatric rocking chair was adapted for a 3-year-old with

a cervical spinal cord injury to allow for weight bearing through the feet with upright

posture and engaged muscle activation in both the arms and the trunk [19]. The

periodicity of motion created by rocking and the requirement of an upright posture to

initiate and maintain the rocking motion equally promote an accessible, self-initiated, and

engaging activity for children of all ages, including those patients who have limited

capacity to move and interact with their environment. Rocking would seem then to

complement and reach beyond clinically delivered, activity-based therapies such as

locomotor training.

Clinically derived improvements in trunk control could be safely practiced by providing a

home-based, age-appropriate, and accessible activity that encourages repetitive activation

of the muscles associated with trunk control. Additionally, enjoyment of an activity is an

important motivator for children which results in increased time/practice engaged in the

activity, and in turn accelerates improvement and lasting use. To study the benefits of

rocking, however, it is necessary to design a rocking chair that meets the particular needs

of this population.

2.10 Rocking Chair Therapy Review

Rocking chairs have been investigated for their effectiveness in several domains,

including postoperative recovery, enhancing physical function in the elderly, workplace

ergonomics, and emotional regulation in substance-use disorder. Investigation of the

effects of postoperative rocking revealed improvement in several measures of recovery

including postoperative ileus duration and [42, 43] and time to first flatus [44].

Research on the use of rocking by elderly people showed improvements in

physical performance as measured by the Berg Balance Scale, maximum knee extension

strength, and maximum walking speed [45]. Other work has shown activation of the

rectus abdominis during rocking, and improved rectus abdominis strength in elderly men

as measured by number of sit-ups completed after six weeks of rocking chair use [46].

Pierce et al. observed a rise in blood pressure among hypotensive older adults during

rocking, which suggests that rocking may improve cerebral perfusion in older adults with

hypotension and thus help to slow Alzheimer’s disease (AD) progression or improve

function of AD patients [47].

Udo et al. examined the effects of rocking on pain and discomfort during seated

work in an office environment, and found a significant decrease in pain in the neck,

shoulders, back and lower back, but an increase in hip pain [48]. Cross et al performed a

study involving veterans with substance use disorder and concluded that vestibular

stimulation through rocking chairs might help self-regulate mood and cravings,

potentially reducing relapse risk [49].

There were no studies found in the current literature that applied or investigated

the use of rocking chairs in the context of pediatric SCI.

2.11 Instrumented Rocking Chair to Monitor Muscle Activation

In addition to providing a way to allow children with SCI to activate trunk

muscles in the home, a rocking chair specifically designed for children with impaired

trunk control and instrumented with sensors, would provide researchers and clinicians

with an opportunity to collect data about the characteristics of the rocking activity, and

such a device may be able to provide information about muscle activation and progress in

trunk control. A quantifiable method of collecting data which could be utilized on an

ongoing basis to assess and track changes in muscle activity would be highly beneficial

to clinicians. Additionally, if data could be collected to allow the tracking of activities

and trunk control in the home, clinicians would have a better picture of patient activity

levels beyond the clinic, as well as an added dimension to the assessment of progress.

2.12 Existing Technology to Activate and Assess Trunk Muscles in

Pediatric SCI

A thorough search of both relevant literature and commercial equipment offerings

does not reveal the availability of rocking chair-based therapy tools for pediatric SCI

patients. Additionally, there are no instrumented devices that have been designed to

capture biometric data for purposes of assessing muscle activation in this patient

population. These gaps in the technology provide a clear opportunity to develop a system

to potentially provide a solution for both consumer and clinical use to extend the benefits

of ABT beyond the clinic, while providing researchers and clinicians insight into muscle

activation enabled by LT.

There are, however, products on the market that look to promote the use of trunk

muscles by individuals with SCI. One example is the TherAdapt Wheelchair Platform

Rocker (TherAdapt Products Inc, Ludington, Michigan, USA), shown in Figure 7 (left),

which is an add-on base to a standard wheelchair that allows a patient to increase their

activity level through rocking [50]. Traditional rocking chairs are also available,

however, as noted previously, these must be modified for use by children with SCI and,

even after modification, significant safety concerns and practical limitations exist.

Other therapy devices designed to provide trunk exercise are available

commercially. One example is the Meerkat stander (Etac AB, Torrance, California, USA)

shown in Figure 7 (right). This device is a standing frame which supports an upright

standing position for children needing varying levels of support. The central column can

be positioned in front of or behind the child and support can be adjusted to meet

Figure 7: TherAdapt Wheelchair Platform Rocker (left), and Meerkat stander (right)

Sources: www.theradapt.com/store/ShowProduct.aspx?ID=118;

www.etac.com/en-us/us/products/pediatrics/standing/r82-meerkat/

individual needs. To encourage movement, the base can be set up with wheels or an

optional rocker base which provides a mechanism to develop posture control [51].

None of the devices reviewed incorporate sensors to provide information to assess

use of the device or track improvements in trunk control.

2.13 Dissertation Organization

Chapter 1 has introduced the project and provided an outline of the specific aims being

pursued.

Chapter 2 has provided an overview of the issues involved with pediatric SCI, trunk

control, and ABT necessary to an understanding of the motivations and techniques

used in this project.

Chapter 3 will cover the methods used to achieve Aim 1 and the outcomes obtained. The

objectives that have been pursued to achieve this aim are:

Objective 1: Design of a rocking chair for children with SCI using a Quality

Function Deployment (QFD) process.

Objective 2: Fabricate the rocking chair.

Objective 3: Experimentally verify safety and operation of rocking chair.

Chapter 4 will focus on the methods used to achieve Aim 2, and describes the activities

undertaken to achieve the following objectives:

Objective 1: Characterize muscle activation while rocking for typically developing

children and children with impaired trunk control.

Objective 2: Describe the differences between muscle activation for

typically developing children and those with impaired trunk

control.

Chapter 5 will focus on the methods used to achieve Aims 3 and 4, and describes the

activities undertaken and results obtain in the completion of the following

objectives:

Objective 1: Characterize the changes in 1) applied forces on footrest, seat and

armrests, and 2) kinematics such as frequency, amplitude, and

accelerations, when children with different trunk control

capabilities rock.

Objective 2: Produce a model to predict patient’s trunk control capabilities.

CHAPTER 3

ROCKING CHAIR DEVELOPMENT

Introduction

The methodology chosen for the design of the rocking chair for children with SCI

is the quality function deployment (QFD) design process, a quality system that was

originally developed in Japan in the 1960s and has now been adopted by major

corporations worldwide. It is used to translate the needs of the end user (often referred to

as “needs of the customer”) into specific design characteristics and to implement these

characteristics in the design of the finished product [52]. The QFD process begins by

identifying the needs that the design should meet. This is accomplished by gathering

information from stakeholders about the specific needs and priorities for the product. This

information is then used to develop specific design features and requirements, which are

prioritized using tools like the House of Quality (HOQ) matrix to link them to the

importance of each of the needs identified earlier. Multiple alternative design concepts

that address the prioritized design requirements are developed and evaluated for how well

each alternative fulfils the identified needs. A detailed design is then developed based on

the selected concepts and drawings are created, materials are specified, and the

manufacturing processes are determined. Based on this design, a prototype is fabricated

and tested. Test procedures are designed to evaluate how well the prototype meets the

needs originally identified. This helps to validate the design, identify any potential issues,

and gather feedback for further improvements. The design may then be iterated until any

identified issues have been addressed. This design method can also be extended into the

manufacturing and distribution of the product to ensure that the manufactured product

continues to meet the standards set by the prototype build, and that it actually meets the

needs of the end users [53].

The aim of this study is to apply these engineering design methodologies to the

design, fabrication, and evaluation of a rocking chair that will meet the needs of children

with SCI and will allow for further study of the effects of rocking in this population.

Materials and Methods

3.2.1 Design Process

The initial phase of the QFD methodology involved identifying the fundamental

needs that the rocking chair should address. To accomplish this, a focus group discussion

was conducted with five therapists experienced in working with children in the target

demographic. Prior to the discussion, important topics to cover were identified, and a list

of questions designed to elicit information about needs the chair should meet was

produced. During the session, open-ended discussions were encouraged to allow

participants to identify needs without biasing them towards any predetermined

conclusions. As discussion progressed, the list of topics and questions was referenced to

ensure that all important issues had been addressed, and any pertinent questions were

asked. Conversation moved from general needs to more specific ones, and finally to the

suggestion of specific features to meet the identified needs. The needs identified during

the meeting were captured through notes taken on a pen board, and through audio

recordings which were reviewed later to ensure that no needs had been missed. The

identified needs were then compiled into an online form which the therapists filled out to

rate the importance of each need on a scale of 1 to 10. This ranked list of needs was used

to develop potential features and design requirements for the rocking chair.

When considering the design and setup of the rocking chair, it was also important

to take the anthropometric characteristics of the intended users into consideration.

Anthropometric data was collected from 22 children in the target population and was

used to define design targets that would ensure that the rocking chair would fit the

intended population of users.

The primary anthropometric measures considered in defining rocking chair design

targets were popliteal height (PH), buttock to popliteal length (BPL), and weight. PH is

the measurement from the bottom of the foot to the back of the thigh just above the knee

while sitting [54], and the range of PH in intended users is closely related to the range of

adjustments that can be made to the footrest height. BPL measures the length from

behind the buttocks to the back of the lower leg, just below the knee while sitting [54],

and helps to determines the required seat depth. Seat depth is measured from the front

edge of the seat to the backrest, and too deep of a seat can interfere with free movement

of the legs during rocking, and so is specified in terms of the maximum allowed depth.

These needs, features and design requirements were incorporated into a “House of

Quality” (HOQ) matrix which provides a structured approach for correlating customer

needs with specific design features. Based on the results of the HOQ matrix, a design was

created incorporating the features deemed important for addressing the needs of children

with SCI, and solid models, accurately representing real-world geometries for all

individual rocking chair parts, were generated using SolidWorks (v2020, Dassault

Systèmes SolidWorks Corporation, Waltham, MA). An assembly using these parts was

created to study chair dynamics and to perform stress/loading analysis.

3.2.2 Simulation Using the Finite Element Method

To assess the design, several simulation techniques were employed to validate its

performance. Motion studies that incorporate real-world physics were developed in the

virtual computer model to evaluate the rocking motion and stability of the chair. To

evaluate the structural integrity of the design, finite element analysis (FEA) was also

performed (SolidWorks Simulation V2020). For wood-based furniture products,

Eckelman recommends that the ultimate strength of wood members be reduced by 2/3 to

account for manufacturing variations such as defects and humidity [55]. The mechanical

properties of Baltic Birch plywood are anisotropic and therefore depend on the

orientation with respect to surface grain. For safety, the lowest reported values

(perpendicular to the face grain) of 3.58 x 107 N/m2 for tensile strength, and 2.48 x 107

N/m2 for compressive strength were used [56]. In the FEA analysis, these values were

reduced to 1/3 of the rated strength, or 1.19 x 107 N/m2 for tensile strength and 8.27 x 106

N/m2 for compressive strength. The factor of safety (FOS), which is a ratio between

ultimate stress of the material and the maximum working stress expected to be seen

during normal use, was used to evaluate the design at each iteration. An FOS of 1.5 is

typically used when evaluating reliable materials in conditions that are not severe (i.e.

high temperatures, wet conditions, etc.) [57].

To simulate a worst-case scenario (an individual much larger than the intended

user of the chair) the model was loaded with 136 kg (300lb) applied to the seat of the

chair, and factor of safety and deflection plots were generated.

Fabrication

Parts for the rocking chair were cut from Baltic birch plywood on a Computer

Numerical Control (CNC) router (ShopBot Tools, Inc. Durham, NC), with tool paths

generated based on the validated solid model. Other miscellaneous parts and hardware

(nuts, bolts, etc.) were purchased from commercial sources, and the chair was assembled,

wood parts were sealed with lacquer, and painted with ocean scenes.

Rocking Chair Mechanical Testing

3.4.1 Tipping

The base of the chair was blocked to allow the chair to tip instead of sliding on

the floor when a force was applied. The chair seat was then loaded with plate weights

(59 kg [130 lbs.], the maximum expected weight of patients in the target population), and

a horizontal force was applied at the top of the seat back. To apply and quantify this

load, a rope was tied to the top of the backrest of the chair, and a luggage scale

(American Tourister Travel luggage scale, 38 kg capacity - Model# AT97-635-027-61)

was attached. The scale was then pulled horizontally in the desired direction until the

chair reached the point at which it would continue to tip over (“tipping point”) if not

restrained. The maximum force exerted, and the lift height of the lifted side at the tipping

point were recorded, and the angle at which the tipping point was reached was

calculated. This was repeated for five repetitions each in forward, backward and side

loading configurations.

3.4.2 Maximum Load Testing

The ability of the chair to support greater than the maximum expected load was

also evaluated. For this test, 136 kg (300 lbs) of weights (representing 2.3 times the

maximum expected load) was placed on the seat of the chair and the chair was rocked

gently. The chair was then examined for any evidence of damage or weakness such

as cracking or bending of wood elements or fastener pull-out or deformation.

3.4.3 Strength and Stability Testing

Chair strength and stability in the expected range of dynamic loading were tested

with loads of 11 kg (25 lbs), 35 kg (78 lbs), and 59 kg (130 lbs). For these tests, weights

were placed on the seat of the chair and the chair was rocked manually to reach the

maximum possible travel amplitude (the safety stops included in the build effectively

limited maximum travel). During rocking, observations were made to identify signs of

weakness or instability; if tipping was observed, the height that the feet lifted off the floor

was recorded, and the angle of tilt was calculated. After loaded rocking, the load was

removed, and the chair was thoroughly inspected for any signs of damage or weakness as

mentioned previously.

Results

3.5.1 QFD Results

The QFD design process identified needs which fell into four categories. The

highest rated category identified by the therapists was safety, followed by therapeutic

needs, practical and aesthetic aspects, and a need for data to characterize and track the

child’s rocking. Table 2 and 3 show the categorized needs along with the importance

rating each need was assigned and design features proposed to meet that need.

Table 2: Customer requirements and features determined from focus group with physical

therapist (Safety and Therapeutic Need). Ratings on a scale of 1 to 10 and are averaged

for each category of needs.

Customer Needs Rating Proposed Design Features and Testing

Safety

Stability/strength 10.0 Baltic birch plywood construction; comprehensive strength/safety

testing

Secure child in the chair 10.0 Belt/Harness to secure child in seat

Resist tipping 9.8 Glider Style rocker with stable base; safety stop to limit range of

motion

Protect skin 10.0 Padding on seat, seat back, and armrests

Ability to lock rocker 8.0 Mechanism to prevent rocking during transfers etc.

Therapeutic Needs

Enable independent

movement

10.0 Means to position strap for adequate support while allowing as

much trunk motion as possible

Provide support 9.8 Full back, armrests, and straps to secure and support child

Engage leg muscles 9.8 Footrest attached to base to allow children to push with legs

Engage arm muscles 8.3 Armrests for child to push/pull on

Engage Trunk muscles 9.5 Removeable footrest; footrest attached to seat rather than to base

Increase vestibular

stimulation

4.0 Adjustable mode of rocking

Increase sensory

stimulation

6.3 Adjustable mode of rocking; Rocking inherently induces sensory

responses from multiple sense modalities (skin, muscle stretch

sensors,

vestibular

etc.)

Several features were chosen to meet the need for safety. These include the

implementation of safety stops to restrict the chair’s range of motion, padding to prevent

chafing and skin breakdown, the utilization of straps to securely hold the child in the seat,

the mitigation of potential pinch points, and a wide, stable base to prevent tipping.

Table 3: Customer requirements and features determined from focus group with physical

therapists (Practical Aspects/Aesthetic Appeal and Data for Assessment). Ratings on a

scale of 1 to 10 and are averaged for each category of needs.

Customer Needs Rating Proposed Design Features and Testing

Practical Aspects/Aesthetic

Appeal

Sized for target population

(age 1-12; up to 59kg)

8.8 Demographics of target population consulted when sizing chair

features (seat width, depth, footrest location and adjustment

range)

Adjust to fit child 9.8 Adjustable footrest; interchangeable seats; adjustable seat

position (larger seat)

Fit into home aesthetics 8.3 Seats made from commercially available child-sized chairs

Attractive to child 8.5 Painted with attractive ocean scenes

Ease of Transport 8.3 Built in wheels; handles for pushing/carrying; mechanism to

prevent rocking

Useable in multiple contexts 6.5 Detachable work surface; mechanism to prevent rocking for use

in other contexts

Fun for the child 9.3 Adjustable modes of rocking to determine which is most

enjoyable/most effective; ocean scenes painted on rocker to

stimulate child's imagination

Ease of use 9.3 Make similar to traditional rocking chair. Straps to hold feet on

footrest

Data for Assessment

Capture Rocking

Characteristics

6.3 Sensors to measure forces applied to chair surfaces

Muscle Activation Assessment 7.8 Data analysis algorithms to find correlations with muscle

activation

Trunk control Assessment 7.5 Data analysis algorithms to find correlations with trunk control

User-friendly display of data

7.5

Wi-Fi link to sensor data/LabVIEW data capture software

Features selected to address therapeutic needs include a full chair back, armrests,

and moveable straps to secure and support the child’s pelvis and trunk while also

allowing for as much independent movement as possible. Other features selected to

facilitate muscle activation by the child include a footrest to allow the use of leg muscles

in rocking and, a feature allowing for adjustment of the rocking motion to explore the

effects this has on patterns of muscle activation and sensory response.

Practical and aesthetic needs were largely concerned with making the chair easy

and enjoyable to use. To meet the need for the chair to fit the target population (age 1-12;

up to 59kg), the footrest was made adjustable, and two commercial rocking chair seats of

different sizes were modified to attach interchangeably to the base of the chair, and to

adjust forward and back. The need to appeal to children in the target population was met

partly by the inherent appeal of rocking, but also by painting the base of the chair with

attractive ocean scenes to stimulate the children’s imagination.

To meet the need to track the child’s use of the rocking chair, a position sensor,

which tracks the motion of the seat during rocking was added to the rocking chair to

make it possible to record the amplitude and duration of rocking. Force sensors were

also added to the footrest, seat, and armrests to detect the forces applied to the rocking

chair during rocking.

3.5.2 Final Prototype Design

The final prototype design incorporating the selected features was based on a

glider style rocker, which uses linkages to suspend the seat of the chair from a base which

remains stable on the floor. Figure 8 shows the solid model of the design which was used

for motion studies and finite element analysis. Motion study results from early designs

identified a tendency for the seat of the rocker to not return to center after rocking. This

was corrected by adjusting the length and position of the linkages, until further motion

study results showed it to consistently return to center.

Figure 8: Rendered solid model of rocking chair design

As shown in Figure 9, results from the finite element analysis of the final design

indicate that the minimum FOS, which occurs in the front linkage arms, is greater than 5

indicating that the chair is able to support at least five times the applied load of 136 kg

before failure. The maximum reported deflection is 0.024 mm.

Figure 9: Finite Element Analysis results; A - Factor of Safety plot showing that

minimum FOS is >5; B – Deflection plot showing that maximum deflection is less

than .024 mm

3.5.3 Anthropometric Design Factors

Several design factors were directly related to the anthropometric characteristics

of the intended user population. The target specifications for each of these, and the actual

dimensions used in the final design are included in Table 4 along with design notes

about how the specifications were identified.

Table 4: Anthropometrics Driven Design Specifications

Design Targets vs. Actual

Dimensions Design notes

Small seat

max

depth

Design Target:

19 cm

Based on anthropometric data, to accommodate smallest children

without need for padding behind back.

Smaller seat should not be deeper than this.

Actual:

14 cm

Exact seat dimensions depend on the size of commercially available

children’s chairs that are compatible with chair base design.

Large

seat depth

Design Target:

36 cm

Target depth for large seat based upon buttock-popliteal length

data from the target population.

Actual:

38 cm

Exact seat dimensions depend on the size of commercially available

children’s chairs that are compatible with chair base design.

Footrest

to seat

range

Design Target:

23 – 40 cm

Based on anthropometric data (popliteal height) from target

population.

Actual:

22 – 34 cm

Larger range would require greater overall chair height,

which conflicts with need for chair stability.

3.5.4 Prototype Chair

Figure 10 shows selected rocking chair features, and stages in the build process as

well as the completed chair.

Selected materials for the chair build were as follows: Wooden parts were cut

from Baltic birch plywood and assembled using 10 x 3½ inch wood screws (Hillman

Deck*Plus) for all wood-to-wood connections. The seat portion of the chair was

suspended on four equal linkage arms (239mm x 64mm) with ½” ball bearings (R8ZZ

Shielded Bearings 1/2 x 1-1/8 x 5/16 Inch Ball Bearings) that were press fit into

appropriately sized cut-outs and captured with washers attached with wood screws as

shown in Figure 3D. The linkage arms were then attached to the base and chair with ½”-

13 bolts and hex nuts. A position sensor constructed of a rotary potentiometer (TT

Figure 10: Rocking chair design details; Completed rocking chair (A); Painted base of

rocking chair(B); Detail of footrest adjustment locking mechanism (C); Detail of

bearing captured with washer in linkage arm, and revised safety stop (D).

Electronics, P160KNP) was installed on the base of the chair and coupled to one of the

chair’s linkage arms via spur gears. Two commercially available typical rocking chair

seats of different sizes (e.g., ECR4Kids Classic North American Oak Wood Rocking

Chair) were purchased and threaded studs (M6 x 50 mm) were added to the bottom of the

seat to enable attachment to the rocking chair base. Three vertical slots were cut in the

front of the chair base, with a ½”-13 bolt located in the middle slot to secure the footrest

in position. A plastic knob with threaded insert (Morton Glass Fiber Polyamide Multiple

Lobe Knob, Fluted Rim, Threaded Hole, ½”-13 Thread Size) was used to tighten the bolt

and secure the footrest in place. Wooden rails attach to the back of the footrest ramp in

the adjacent outside slots and serve to keep the footrest aligned correctly. Total cost for

all materials used and fabrication of parts on the ShopBot CNC router was kept under

US$ 400.

Safety Testing Results

3.6.1 Tipping Results

Tipping study results show that a minimum of 6.4 kg of force in the horizontal

direction at the top of the seat back is required for the chair to reach the tipping point.

This minimum force occurs at the lowest loading of 11 kg, and higher loads required

greater forces to tip. Overall, less force was required to tip the chair forward than to tip it

back or to the side. The force required to reach the tipping point at each load level for

each axis, as well as the lift height of the lifted side at the tipping point are shown in

Table 5.

Table 5: Chair tipping test results.

Chair loaded with 11, 35, or 59 kg (n=5 trials)

Mean tipping force (kg)

Mean lift height at tipping point (cm)

Tilt Direction 11 kg 35 kg 59 kg 11 kg 35 kg 59 kg

Forward 7.1 11.1 15.0 41.3 31.5 29.8

Back 9.1 14.7 17.4 42.5 33.6 28.8

Sideways 7.7 12.6 18.0 34.3 27.3 26.5

3.6.2 Maximum Load Results

While loaded, the chair was observed for stability, deflection of weight bearing

members, and structural integrity. No sign of weakness or instability was observed during

testing. After removing all weights from the chair seat, the assembly was inspected, and

no structural damage was observed.

3.6.3 Chair Strength and Stability Results (Dynamic Loading)

During dynamic testing, the original safety stop design was found to allow the

chair to bypass the rear stop and capsize at the 59kg (130 lbs) loading level. Based on this

result, the safety stop was redesigned, and the test was repeated with the new safety stop,

which performed well in all tests. No other signs of structural damage or weakness were

observed. At maximum rocking amplitudes, the chair tipped slightly on impact with the

safety stop at the maximum limit of travel. Comparing the angle of tilt during dynamic

rocking to the tipping point angle during tipping tests, the maximum angle of tilt during

dynamic testing was less than 1/3 of the tipping point angle that would be required to

cause the chair to overturn. This adds confidence that the chair will remain stable during

normal rocking. The full array of tilt angles observed during dynamic testing and their

relation to the tipping point are included in Table 5.

Table 6: Chair tilt angle during dynamic loading.

Load Tilt Direction Mean

dynamic tilt*

(degrees)

Maximum

dynamic tilt

(degrees)

Mean tipping

point†

(degrees)

Tipping point

safety factorŧ

11 kg Forward 5.3 6.4 33.4 5.3

Backward 2.1 2.8 34.6 12.2

35 kg Forward 5.5 6.9 24.8 3.6

Backward 4.6 5.0 26.6 5.3

59 kg Forward 5.5 6.7 29.8 4.4

Backward 3.3 3.7 22.6 6.1

*n=10; †n=5; ŧ Tipping point safety factor represents the ratio of tipping angle to maximum dynamic tilt

angle at a given load level.

Discussion

This project aimed to apply engineering design methodologies to develop a

rocking chair specifically for children with spinal cord injury (SCI) to promote trunk

control. Throughout the process, the needs identified in the QFD design process

were used to guide design choices, and to evaluate the final design.

Safety was the highest rated need category and was one of the primary reasons for

designing a custom rocking chair rather than simply adapting a commercially available

rocker. The choice of a glider type rocker, rather than a more traditional rocking chair,

was made primarily to provide a stable base which would resist tip-over. As shown in

Figure 11, it is possible to limit the possible range of motion of this type of chair so that

the center of gravity cannot move outside of the footprint of the base. This makes the

chair naturally resistant to tipping.

Further evidence that the chair will resist tipping was provided by the results of

physical testing which showed that at maximum amplitude of rocking under various

loads, the chair does not come close to the tipping point. As seen in Table 5, the

minimum tipping point factor of safety, defined as the ratio of the tipping point angle to

the maximum dynamic tilt angle, is 3.6. This indicates that the chair would have to tip at

an angle more than three times greater than what is actually seen in dynamic testing

before tipping over. Other physical testing showed that the chair was capable of

supporting the desired loads with an ample factor of safety.

The glider type rocker also has advantages with respect to meeting therapeutic

needs. Independent movement, sensory stimulation, and muscle activation in arms, legs

Figure 11: Rocking chair center of mass At maximum forward and back rocking,

center of mass (blue arrow) stays within footprint of rocking chair and no overturning

moment is created.

and trunk are all identified needs. The large, stable base of the glider type rocker provides

a place to mount a footrest. Patients who have some leg control will be able to use the

footrest to move the chair, providing a way to activate leg muscles. For patients without

leg control, foot placement on the footrest will allow legs to flex as the chair rocks,

stimulating sensory input to the spinal cord from the legs and feet.

Several of the features contributed towards meeting multiple needs. For instance,

in addition to helping meet therapeutic needs, the footrest was also made adjustable to

help meet the need for the chair fit children in the target population. Similarly, straps that

could be adjusted higher and lower on the child’s trunk helped to meet needs for safety,

while also allowing as much independent trunk movement as possible.

The prototype rocking chair is intended for use in further studies in children with

SCI to characterize muscle activation during rocking. As such, some features intended

primarily for convenience in home use were not implemented in the current build but will

likely be implemented in future version of the chair. These include features such as a

mechanism to prevent the chair from rocking during transfers, wheels and handles to aid

in transport, and a detachable work surface for use during other activities.

Limitations

There are several limitations to this work. Lack of human testing in the target

population limits the ability to validate that the design meets the defined needs when

used by children with SCI. Also, until human testing has been completed it will be

unclear what clinical relevance rocking has for these children.

Future Work

Plans for future work include studies with children with SCI to validate that the

chair meets the identified needs when used by the target population. One need in

particular which could not be evaluated without having children rock in the rocking chair

was the need for an activity that is enjoyable for children with SCI. This need was

emphasized by therapists as one of the most important for children in this age group since

enjoyment of an activity leads to motivation to engage in that activity, additional time

spent in the activity, and better results from the activity. Additionally, efforts will be

made to characterize muscle activation patterns in arms, legs and trunk during rocking to

determine if rocking is an effective means of stimulating activation in muscles involved

with trunk control.

Conclusion

This prototype rocking chair provides an opportunity to safely study rocking as a

means of improving trunk control in children with spinal cord injuries. Further testing

and refinement of the design, as well as the incorporation of embedded sensors to

provide feedback on how the chair is used all hold potential for extending its usefulness.

Ultimately, rocking may evolve into a valuable tool for therapists seeking a safe,

accessible means for children with SCI and impaired trunk muscles to safely rock and

activate their trunk muscles.

CHAPTER 4

MUSCLE ACTIVATION DURING ROCKING

4.1 Introduction

Spinal cord injury (SCI) is a debilitating condition that significantly impacts

quality of life. Children with SCI in particular face numerous physical, emotional, and

psychological challenges that often impede their overall development and well-being

[25]. SCI is known to cause long term complications including negative effects on

respiration, bladder and bowel control, cardiovascular function and head/trunk control

[11]. Moreover, SCI in children is more likely to result in secondary conditions, such as

scoliosis, respiratory complications, hip dysplasia, and pressure ulcers [11, 28], and the

incidence of these conditions is greater for children injured at younger ages [3, 17, 27,

28].

4.1.1 Activity Based Therapy

With the existing state of the art science currently unable to fully resolve or

ameliorate the various types and severities of paralysis associated with SCI, traditional

rehabilitation approaches focus on compensation for paralysis via assistive devices,

behavioral strategies, and adaptations to the environment to achieve mobility and

function, and to facilitate activities of daily living [58]. However, researchers have

recently demonstrated that activity-based therapy (ABT) is an effective intervention for

at least partial restoration of intrinsic trunk control, as measured by SATCo score, in

children with SCI [33]. The purpose of ABT is to activate the neuromuscular system

below the level of injury, and to promote the restoration of neuromuscular capacity. ABT

interventions include locomotor training (facilitated walking on a treadmill) with partial

body weight support (systems that modulate the offset of patient weight during the gait

cycle), which aims to mimic typical standing/walking patterns, reinforce appropriate

alignment of the trunk, pelvis, and lower extremities providing locomotor-specific

sensorimotor input to the neural axis [19, 39]. In addition to training on the treadmill, off-

treadmill interventions utilize the activated neuromuscular system targeting task-specific

motor activities to integrate use of the progressively activated trunk in typical activities

such as reaching, standing, and sit-to-stand [33]. ABT also focuses on home and

community integration by giving children opportunities to utilize their new capacity and

practice in a variety of everyday and functional activities [36].

Community integration considerations create opportunities to develop additional

reinforcing activities that increase practice with new neuromuscular capacities and can be

performed in the home and community. From that perspective, rocking in a rocking chair

was explored in a case report by Argetsinger et al [40]. This study describes the trial of a

pediatric rocking chair adapted for use by a child with SCI as an activity intended to

support gains in trunk postural control outside of the clinic. Rocking in this chair created

a self-initiated activity that promoted an upright posture and engaged trunk and arm

muscles with little to no assistance by caregivers. Validating the effectiveness of this

approach in applying the gains acquired through locomotor training is a critical next step

to inform clinical decision-making.

4.1.2 Study Rationale

To better understand the effectiveness of rocking in the context of pediatric SCI,

this study aims to answer the following research questions: 1) Does rocking in a rocking

chair activate trunk muscles in children with SCI and in typically developing (TD)

children? 2) What are typical temporal muscle activation patterns during a rocking cycle

in children with SCI, and how do these patterns differ from those seen in TD children? 3)

Is there a correlation between intrinsic trunk control, as measured by SATCo, and trunk

muscle activation during rocking?

In addition to these specific technical questions, it is important to investigate

practical factors that impact the rocking experience of the user. These factors include the

child’s enjoyment of rocking, and the safety and functionality of the rocking chair

prototype.

4.2 Materials and Methods

A thorough examination of the concept, design methodology, fabrication, and

safety testing of the rocking chair have been described previously [59]. Briefly, to

develop new technology for accomplishing the goals of this study, a custom rocking chair

specifically designed for children with spinal cord injuries was fabricated. Based on a

glider style rocking chair to ensure stability and safety, the base of the chair was

assembled from parts fabricated using Baltic birch plywood, a well-known furniture

material which provided both durability and cost benefits and promoted ease of chair

assembly. Seats from commercially available pediatric-scale rocking chairs of two

different sizes were adapted to attach interchangeably to the custom base to

accommodate children of varying sizes. An adjustable height footrest was also included

in the design of the rocking chair

base to accommodate taller children.

To quantify the amplitude of rocking and to enable tracking of the temporal

location of muscle activation in the rocking cycle, a custom chair angle/tilt sensor was

incorporated into the design. This included a shafted potentiometer configured so that

the rotational position of the shaft was adjusted by the angular movement of one of the

chair suspension arm bearings during the rocking motion. This sensor generated a

dynamic voltage which, when converted to angle of rotation, accurately reported the

position of the seat as it was dynamically rotated from its neutral position by the user.

4.2.1 Participant Recruitment

The current study was performed under an Institutional Review Board (IRB)

approved protocol for parental/caregiver informed consent and assent for children older

than 7 years (University of Louisville IRB #17.0725). Two cohorts were enrolled:

children with SCI and TD children, based on specific eligibility criteria. Children with

SCI, age 1-12 years and currently participating in locomotor training, recently discharged

(< 1 month) or, if discharged between 1 and 24 months prior, were eligible with a

medical screening to determine health status and/or safety concerns. For TD children, age

1-12 years with no history of SCI or presence of neurological disorder, musculoskeletal

disease, cardiovascular, pulmonary, or respiratory condition that would affect typical

function, and the ability to follow age-appropriate instructions were eligible. For both

cohorts, children with a physical condition or recent illness, identified via medical screen,

that would prevent participation in rocking were excluded.

4.2.2 Participant Characteristics

A total of 11 children with SCI and 10 TD children were enrolled over a 22-

month period. Age and height at time of testing, and sex were recorded for all

participants in both groups. American Spinal Injury Association (ASIA) Impairment

Scale (AIS) [60] was recorded for participants in the SCI group whose age and etiology

allowed it to be determined. Injury level, time since injury, etiology, and SATCo scores

were also recorded for participants in the SCI group. SATCo score is an indication of a

child’s capacity to sit upright in both static and dynamic conditions with a facilitated

neutral pelvis position and varying biomechanical levels of trunk support, and represents

the patient’s inherent, uncompensated level of trunk control [33]. Scores range from 0 to

20, with higher values indicating better trunk control. SATCo was the primary measure

of trunk control impairment and a primary descriptor of the population with SCI

participating in this study.

4.2.3 Anthropometrics and Rocking Chair Fit

When considering the design and setup of the rocking chair, it was important to

take the anthropometric characteristics of the users into consideration. The following

anatomical features were considered to be important anthropometric measurements that

might affect rocking chair fit.

Popliteal Height (PH) is the measurement from bottom of foot to underside of

thigh while sitting. This helps to determine footrest height relative to the top of the seat.

For a stationary chair, the seat to floor distance could match the PH, but for a rocking

chair this leads to the feet lifting off the footrest as the chair rocks back. Consequently,

the footrest should be set up so that, the knees are slightly elevated off the seat when the

chair is centered in the neutral position, and the feet do not lift when the chair rocks back.

Buttock-popliteal length (PBL) is the measurement from behind the buttock to the

back of the lower leg just below the knee while sitting. This determines the depth of the

seat measured from the front edge of the seat to the backrest of the chair (or to padding

placed behind the user’s back). The seat depth should be less than the BPL to allow for

movement of legs during rocking. If the seat is too deep, padding can be added behind

the child, but too much padding may disrupt the balance of the chair by moving the user’s

center of gravity forward.

User weight can interact with seat adjustments that move the center of balance

forward or back due to size factors, such as when a smaller child has padding placed

behind their back, or a larger child has the seat moved back. This is primarily a problem

for larger, heavier children since a smaller child’s weight is usually not enough to change

the balance significantly.

Trunk width does not affect chair motion but should be checked to ensure that the

child can sit comfortably with no squeezing or chafing from the sides of the chair.

4.2.4 Rocking Chair Setup and Fit Adjustment

Table 7 is a table of identified parameters that were considered when seating a

child in the chair, including methods of adjustment for optimal fit and balance. When

setting up the chair, it may be impossible to meet all the fit criteria simultaneously. For

instance, if the seat balance is too far back, and the footrest has already been adjusted all

the way down, it may not be possible to move the child forward to correct the balance

without moving them too close to the footrest.

Table 7: Chair Setup and Fitting Parameters

Fitting

parameter Definition and adjustment

Seated

balance

When sitting still in the rocking chair, the seat should not tilt more than 10 degrees in

either direction from the neutral position. This can be compensated for by adding padding

behind the child to move them forward, or by sliding the seat back or forward, or by

switching to the larger/smaller seat.

Footrest

placement

When the child rocks back in the rocking chair, the feet should not lift off footrest. This

means that in the neutral position, there will be a gap between the front edge of the top of

the seat and bottom of the child’s leg just above the knee. The size of the gap will vary

depending on the chair setup and child size, so the most reliable way to check this is to

watch the child rock and adjust the footrest position until the bottoms of the legs (behind

the knee) just touch the top of the seat when the chair reaches its farthest-back position,

and the feet do not lift off the footrest.

Seat depth When the child rocks all the way forward, the backs of the calves should not bump the

front of the seat. If they do, switch to the smaller seat, or add padding behind the user to

decrease the effective seat depth.

Seat width When seating the child, the width of the chair should be checked to ensure that it is not

too tight. If the sides of the seat squeeze or rub against the child significantly, the larger

seat should be used if possible. Other width adjustments are not possible with the current

design

Child

weight

The chair has been tested and is safe for up to 130 lbs. Children heavier than this should

not use the chair.

4.2.5 Chair Functionality Assessment

To validate the basic operation of the chair, a TD child first evaluated the chair by

sitting in the seat to test its balance, followed by rocking to assess both comfort and

function. Once basic chair operation was verified, children with SCI were allowed to rock

in the chair. Children were asked to rock in several different ways to evaluate which

muscles they were capable of using to rock the chair: 1) Free rocking, allowing

participants to rock in any manner of their choosing; 2) arms-off-armrests, to promote use

of trunk muscles; 3) leg-only rocking, restricting trunk motion to assess the contribution

of leg muscles to rocking; 4) footrest removed, so that only upper body muscles

contribute to rocking. Results from these modes of rocking were noted and compiled to

better understand muscle use strategies.

During rocking sessions, children were observed closely while technicians

engaged with them in conversation to understand their experience. Enjoyment of rocking,

difficulties with making the chair rock, methods of initiating and maintaining rocking,

and any potential safety or practical issues were observed and noted. Additionally,

parents, children, engineers, and therapists engaged in debriefing conversations after

each rocking session to obtain reactions and feedback, and to suggest possible

improvements to the chair and protocols. Videos of rocking sessions were also reviewed

to ensure that all data pertaining to chair use had been captured.

4.2.6 Surface Electromyography (sEMG) Data Acquisition

sEMG signals from muscles in the arms, legs, and trunk of both TD and SCI

groups were recorded using wireless electrodes (Cometa Pico EMG, , pre-amplified,

bipolar , 29 mm electrode spacing, 2000 Hz sampling rate, Cometa SRL, Milan, IT) as

shown in Table 6.

Table 8: Muscles targeted for EMG readings

Trunk Muscles Arm Leg

Cervical paraspinal (PSC) Biceps brachii (BB) Rectus femoris (RF)

Thoracic paraspinal (PST) Long head of triceps (TB) Medial hamstring (MH)

Lumbar paraspinal (PSL) Pectoralis (PEC) Tibialis anterior (TA)

Rectus abdominis (RA) Medial gastrocnemius (MG)

Oblique (OB)

Appropriate locations on the skin above the muscle of interest were prepared by

cleaning with alcohol swabs, then electrodes were placed over each muscle while the

child was seated on a stationary table (see appendix D for full electrode placement

protocol). Subjects were then asked to sit in (TD) or were placed in (SCI) the chair

(example in Figure 12), a safety strap was placed around the waist, and additional straps

placed higher on the trunk if necessary for support.

Each child was instructed to sit quietly for one minute while sEMG was recorded

to establish baseline muscle activation. All children were then instructed to rock in the

chair using whatever methods were most natural to them. Once the child was accustomed

to rocking and was observed to be rocking in a regular rhythm, sEMG was recorded for

one minute. This process was repeated for the other three modes of rocking. Appendix D

includes an outline of the full data collection protocol.

sEMG data was acquired at 2000 Hz and imported into MATLAB. sEMG data

offset was removed by subtracting the mean value of all data points from each data point,

Figure 12: A participant with SCI using the custom rocking chair prototype.Note: strap

around chest for support, and footrest raised to support feet.

and a 300-point moving window root mean square (RMS) was calculated for all datasets

using custom MATLAB scripts (Appendix A).

4.2.7 Rocking Cycle

For this study, a single rocking cycle was defined as the position of the chair as it

swings from the extreme front position to the extreme rear position and back to the front.

Thus, the beginning and end of the cycle is defined as the point at which the chair

reverses direction at the front of the cycle. As shown in Figure 13, the cycle was broken

into two phases: Phase 1 is backward (BW) motion, starting with the chair all the way

forward, and lasting while the chair swings backward to the farthest back point in the

Figure 13: Sample position sensor data from one rocking cycle, to highlight cyclical

phase of rocking Center vertical line marks change in direction; left quadrant is

backward motion; right quadrant is forward motion.

cycle. Phase 2 is forward (FW) motion, starting with the chair at the farthest back point,

and lasting until the end of the cycle with the chair again at the farthest forward point.

Each of these phases was divided into early (EBW, EFW) and late (LBW, LFW) stages.

Accordingly, the chair rocking direction reverses from forward to backward at the end

point of the cycle, and reverses from backward to forward at the midpoint of the cycle.

4.2.8 Artifact Removal in sEMG Data

As is common with sEMG data, many of the trunk muscle data sets were

contaminated with cardiac artifacts. To remove this interference, two methods were used.

For data used to calculate increase in muscle activation while rocking, it was possible to

use the data between heartbeats to calculate the magnitude of the sEMG signal. To delete

the contaminated data, a custom MATLAB script (Appendix A) was used to identify the

interfering signals and delete them from the dataset. Before running this script, each

dataset was manually inspected to determine an appropriate threshold for detecting

heartbeats. Typically, the cardiac signal was strongest in the left pectoral sEMG, so this

signal was used initially to locate the heartbeats, and the start and end points of each

heartbeat were identified. The erroneous data was then deleted from all data streams and

the remaining data was concatenated.

To determine the temporal patterns of muscle activation during the rocking

cycle, it was necessary to use a method of cardiac contamination removal that preserved

the time-dependent nature of the data. Similar to the previous method, this involved

identification of the start and end of each heartbeat. Since there was both significant

variation in the strength of the cardiac signal across full session recordings and variation

between subjects regarding which muscle activation data was contaminated, the data

from each trunk muscle was manually inspected. Muscle activity recordings with

contamination were flagged for correction, and the contaminated data was deleted. A

custom MATLAB script (Appendix A) was written to detect the rocking cycles, and

missing data that had been deleted due to cardiac contamination was replaced by the

averaged data from the same point in the rocking cycles with intact data for that part of

the cycle.

In addition to the cardiac contamination, there were also two other, less prevalent,

sources of data contamination. First, to capture data from the desired muscles, some

sEMG sensor bodies were required to be positioned in locations that brought them into

contact with the seat during rocking, potentially introducing artifacts into the recorded

data in the event of a collision. Second, extraneous movements on the part of the child,

such as a hand gesture, could introduce a signal unrelated to the rocking cycle into the

data. These were identified and removed manually by inspecting the EMG data and

cross-referencing with the corresponding video recording of the data collection session.

4.2.9 Muscle Activation Amplitude Calculation

For each participant, the mean sEMG (RMS) amplitude during rocking was

calculated for each muscle. Baseline amplitude was also calculated by taking the three

second period with the lowest mean sEMG (RMS) amplitude during quiet sitting for each

muscle. Normalized activation amplitude for each muscle was calculated by subtracting

the baseline amplitude from the mean amplitude during rocking of the corresponding

muscle. Normalized activations were then plotted in a heatmap.

4.2.10 Prototype Cycle Creation

To determine the timing of muscle activation within the rocking cycle, an average

or “prototype” rocking cycle was created for each muscle. To do this, a custom

MATLAB script (Appendix A) was used to break the EMG data from each subject into

rocking cycles using the data from the position sensor mounted on the rocking chair.

These activation cycles for each muscle were averaged for each subject, and the resulting

average cycles were normalized to a scale of 0 to 1 using the formula:

𝑃

𝑛

− 𝐶𝑦𝑐𝑙𝑒

𝑚𝑖𝑛

𝐶𝑦𝑐𝑙𝑒

𝑚𝑎𝑥

−

𝐶𝑦𝑐𝑙𝑒

𝑚𝑖𝑛

Eq. 1

where Pn is the nth data point, Cyclemax is the maximum value for the given muscle, and

Cyclemin is the minimum value for the given muscle. The normalized cycles for each

muscle were then averaged for the SCI group and the TD group, to produce typical

activation cycles, and the resulting curves were again normalized to a scale of 0 to 1.

The normalized cycles for each muscle were plotted with SCI and TD data on the

same graph to facilitate comparison.

4.2.11 Statistics

To address the first research question, “does rocking in a rocking chair activate

trunk muscles in children with SCI and in TD children?” the mean sEMG (RMS)

amplitude during rocking was calculated for each muscle activity recorded. Baseline

amplitude was also calculated by taking the three second period with the lowest mean

sEMG (RMS) amplitude during quiet sitting for each muscle. One-sided signed-rank tests

were then used to test the hypothesis that for both SCI and TD groups, sEMG amplitude

(RMS) for each muscle during rocking would be greater than baseline sEMG amplitude

(RMS) for the same muscle during quiet sitting. P-values were corrected for multiple

hypothesis testing using the Holm-Sidak method [61].

To address the second research question, ”What are typical temporal muscle

activation patterns during a rocking cycle in children with SCI, and how do these patterns

differ from those seen in TD children?” cluster analysis was used to investigate how the

timing of muscle activation in the SCI group compared to those observed in the TD

group, and to determine if different rocking styles contributed to different muscle

activation patterns in trunk muscles. To achieve this, the rocking cycle was subdivided

into four segments, as described previously, and shown in Figure 13. EBW (first 25% of

the cycle), LBW (26-50%), EFW (51-75%), and LFW (76-100%). A custom MATLAB

script (Appendix A) was used to calculate the mean normalized muscle activation within

these segments for each subject and each trunk muscle. This data was aggregated and

used to perform cluster analysis. Cluster analysis was performed in MATLAB, using the

‘linkage’ function with the ‘average’ distance method to find clusters of subjects who

used similar temporal muscle activation patterns while rocking. After clusters were

identified, distance between each SCI subject and the TD cluster was calculated using the

formula:

𝐷 = (𝑇𝐷�25 − 𝑆𝐶𝐼25)2 + (𝑇𝐷50 − 𝑆𝐶𝐼50)2+(𝑇𝐷75 − 𝑆𝐶𝐼75)2 +

(𝑇𝐷100 − 𝑆𝐶𝐼100)2Eq. 2

where D equals distance, and TDx and SCIx represent the mean activation for the given

muscle over the respective segment of the rocking cycle. These distances were then used

to determine which SCI clusters were more similar to the TD cluster, and the mean

activations were plotted for the TD subjects and each SCI cluster [62].

Two-sided t-tests were used to test for differences in activation in each segment

between the TD group and each of the SCI clusters, and significant differences were

noted on plots.

To examine the third research question, “Is there a correlation between intrinsic

trunk control, as measured by SATCo, and trunk muscle activation during rocking?”

Spearman correlation analysis was performed to assess the correlation between SATCo

scores and trunk muscle activation amplitude. For purposes of this analysis the two

children with acute flaccid myelitis (AFM) were excluded as their impairment

presentations are atypical, not uniform across spinal cord levels, and often reflect greater

to lesser impairment in a cephalad-caudal direction. In contrast, other participants

demonstrate higher SATCo scores with greater control and activation of lower thoraco-

lumbar muscles. Results were compiled in a table and significant correlations were noted.

All confidence levels were set at 95% (p < 0.05). Prism 10 (GraphPad Software,

San Diego, CA, USA) was used for the analysis.

4.3 Results

4.3.1 Participant Characteristics

Characteristics of participants including age, height, SATCO score, and injury

related details are shown in Table 9.

The SCI group included 7 males and 4 females between 3 and 11 years of age

(mean = 7.1, SD = 2.7), and heights ranging from 77 to 42 cm (mean = 120, SD = 22).

SATCo scores ranged from 0 to 19 (out of 20) (mean = 13.1, SD = 5.8). Two children

presented with cervical SCI, 6 with thoracic SCI, and 1 with lumbar SCI. Time since

injury ranged from 1.5 to 6.5 years (mean = 4.1, SD = 1.7).

The TD group included 6 males and 4 females with ages ranging from 5 to 11

years (mean = 7.2, SD = 2.4), and heights from 111 to 143 cm (mean = 128, SD = 14).

Table 9: Participant characteristics

Participant

Group Age

(yrs)

Sex Height

(cm)

SATCo

Score

AIS Injury

level

Time since

injury (yrs)

MOI Etiology

P216 SCI 7 M 131 0 * - 1.5 NT Acute Flaccid Myelitis

P188 SCI 4 F 94 8 * T2-T5 4 NT Myelomalacia

P23 SCI 7 M 140 11 B C4-C7 4.5 NT Transverse Myelitis

P34 SCI 8 M 128 11 A T2-T8 6.5 NT Spinal Astrocytoma

P32 SCI 9 F 129 12 A T2-T3 6 T Motor Vehicle Accident

P187 SCI 3 M 77 12 * T11 2.5 NT Spinal Cord Ischemia

P204 SCI 9 M 130 15 A L4 2 T Motor Vehicle Accident

P217 SCI 3 M 91 18 * C1-C3 3 NT Perinatal

P15 SCI 9 F 127 19 A T12 5.5 NT Epidural Abscess

P21 SCI 11 F 142 19 A T10 5 NT Spinal Stroke

P215 SCI 9 M 139 19 * - 4 NT Acute Flaccid Myelitis

SCI Mean (SD) 7.1 (2.7) 120 (22) 13.1 (5.8) 4.1 (1.7)

SCI Median (IQR) 7.8

(5.5,9.0)

129

(110,135)

12.0

(11.0,18.5)

4.1 (2.8,5.2)

SCI Range 3-11 77-142 0-19 2-7

P218 TD 9 M 143 - - - -

P38 TD 8 F 129 - - - -

P219 TD 5 F - - - - -

P220 TD 5 M 115 - - - -

P221 TD 8 M 134 - - - -

P222 TD 5 F 111 - - - -

P223 TD 11 M 149 - - - -

P224 TD 9 M 141 - - - -

P225 TD 5 M 119 - - - -

P226 TD 5 F 113 - - - -

TD Mean (SD) 7.2 (2.4) 128 (14)

TD Median

(IQR)

6.5

(5.2,9.0)

129

(115,141)

TD Range 5-12 111-150

*AIS not valid under 5 years of age [63]; SCI = Spinal cord injury; TD = Typically Developing; SATCo = Segmental Assessment

of Trunk Control; AIS = ASIA Impairment Scale score; MOI = Mechanism of injury; T = traumatic; NT = Non-traumatic

4.3.2 Safety, Functionality, and Enjoyment

Data gathered during the observation of rocking sessions are summarized in Table

10.

Fit: Three of the children in the SCI group had minor issues with fit that were

addressed satisfactorily by adding padding to the footrest or behind their back. Two other

children had fit issues that couldn't be fully corrected. One was too big for the chair and

the balance was off towards the rear (participant P21 in Table 10 for details of chair

adjustments). The other was too small for the chair even with padding on both the

footrest and back of the seat (participant P187 in Table 10). Both these children were

still able to rock, but their rocking mechanics (how they rock, and their overall

experience rocking the chair) may have been affected by the balance and fit of the chair.

Safety: No adverse safety events or injuries were experienced during trials with

children in the rocking chair. It was however, observed that a cross brace under the

seat could create a potential pinch-point for smaller children.

Functionality: Regardless of previous experience with rocking or degree of

impairment, all children in both groups were successful at making the chair rock. The two

youngest children (both three years old) had some difficulty timing their efforts but were

able to make the chair rock. After some practice, their timing improved, but the amplitude

of rocking remained well below the chair’s full range of motion.

All children in both the SCI and TD groups were able to fit in the rocking chair.

The interchangeable seats and adjustable footrest accommodated most children without

alteration, but for those with shorter legs, it was necessary to insert padding between the

child and the seat back and, in one case, to place foam blocks on the footrest to enable

their feet to reach comfortably. This adjustment allowed them to engage in rocking but

made it more challenging to position them correctly.

Debriefing discussions following rocking sessions provided suggestions for

functional improvements to the rocking chair. These suggestions included a larger

footrest with straps and a non-slip surface to keep feet in place, and a mechanism to

prevent rocking during transfers or other activities.

Enjoyment: During conversation several of the children specifically said they

enjoyed rocking in the rocking chair, and two of the children, who had to return for a

second session due to problems with data collection, both expressed their eagerness to

return. Another indication of enjoyment was that many of the children rocked

spontaneously while not specifically engaged in a rocking task, and in some cases, it

was difficult to convince them to sit still.

Table 10: Participant experience while rocking

Participant

ID Description

Nervousness

Discomfort

Trunk

Movement

Leg

Use

Seat

Size

Footrest

Height

(cm)

Added

back

padding

(cm)

Added

seat

padding

(cm)

P15 Initially nervous, but confident by end of session Y N None Y S 15 0 0

P21 Very focused on completing rocking tasks; said she might

use the chair at home if she had one, and her mom used to

rock her when she was sick

N N Counter Y L 0 5 2

P23 Pretended to be in rocking boat; asked many questions;

spontaneous rocking between trials, but a little tired by end

N N Sync N S 15 0 0

P32 Focused on tasks; Helped with EMG placement N N Sync N S 0 0 0

P34 Commented rocking was fun; spontaneously rocked

between trials

N N Sync N L 7 0 0

P215 Enjoyed pushing with legs. Said seat was too narrow N Y None Y S 0 0 0

P187 Difficulty timing efforts, but made chair rock, and

improved over course of session

N N Counter N S 24* 8 0

P188 Vigorous rocking, but tired by end of session N N Sync Y S 15 7.5 0

P216 Returned for 2nd session, and parent mentioned they had

purchased a rocking chair for home use

N N None Y L 0 4 0

P217 Played with toys during rocking N N Counter N S 15 10 0

P204 Returned for 2nd session due to data collection error, and

parent commented “he couldn’t wait to come back here and

do this”; child said he would rock at home if he had a

rocking chair this cool

N Y Sync N L 0 0 0

*Included 9 cm block on top of footrest; Y= Yes, N= No, Sync = Movement synchronized with rocking so weight shifts forward during

forward chair movement; Counter = Trunk movement counter to rocking so weight shifts back as chair moves forward; S = Smaller seat; L

= Larger seat

There were no major, negative reactions to the rocking chair. One of the children

did initially express some apprehension that the chair would tip over but was reassured

when the safety stops which limit chair tilt were demonstrated, and quickly gained

confidence as they rocked. Two of the children rocked vigorously at first but expressed

that they were feeling tired toward the end of the rocking session.

4.3.3 Observed Trunk Motion During Rocking

The majority of children in the SCI group moved their trunk and/or head in time

with the rocking cycle. Of these, five leaned their head and upper trunk, shifting their

weight forward while the seat moved forward, and shifting their weight back while the

chair moved back. In contrast, three children instead moved their trunk in a

counterbalance rocking strategy, leaning forward from the waist as the seat of the chair

moved back. This effectively kept the center of mass of the seat/child in approximately

the same place but moved the seat back and forth. Finally, three children maintained a

mostly stationary trunk.

4.3.4 Increase in Muscle Activation During Rocking

Mean RMS muscle activation during rocking was significantly higher than during

quiet sitting (p < 0.05) for all twelve muscles tested in both SCI and TD groups, see

Table 9 and Table 10 respectively.

Table 11: Increase in muscle activation while rocking for children with SCI

Muscle n

Baseline:

Median[IQR]

Rocking:

Median[IQR]

Median Increase

(CI*) p-value†

Cervical PS‡8 6.40[4.39, 13.2] 14.1[11.0, 35.0] 7.14(3.92 to 27.0)* 0.006

Thoracic PS 11 5.60[3.59, 9.08] 18.3[12.8, 37.7] 12.8(1.46 to 32.8) 0.006

Lumbar PS 11 3.78[3.40, 5.60] 9.49[5.43, 21.5] 6.13(0.56 to 21.1) 0.006

Pectoral 11 5.15[3.37, 10.7] 10.8[7.79, 19.8] 4.49(1.14 to 19.6) 0.006

Biceps 11 4.01[3.24, 5.66] 16.0[7.30, 30.1] 13.0(3.46 to 79.1) 0.006

Triceps 11 3.37[3.23, 6.20] 18.0[10.5, 34.3] 14.2(6.50 to 31.2) 0.006

Rectus Abdominis 11 3.58[2.92, 3.82] 6.76[3.57, 14.3] 3.56(0.45 to 15.4) 0.006

Oblique 11 4.01[3.28, 4.64] 10.0[4.79, 13.3] 3.82(1.00 to 10.6) 0.006

Rectus Femoris 11 3.13[2.81, 3.26] 3.84[3.42, 13.4] 0.71(0.42 to 25.1) 0.006

Hamstring 11 3.21[2.85, 3.31] 4.45[3.78, 13.5] 1.29(0.50 to 21.9) 0.006

Tibialis 11 3.17[2.75, 4.19] 3.54[3.14, 18.0] 0.44(0.37 to 8.18) 0.006

Gastrocnemius 11 2.98[2.70, 3.88] 3.69[3.22, 9.68] 0.86(0.36 to 10.1) 0.006

All muscles tested showed a significant (p < .05) increase in muscle activity (as measured by mean of

EMGRMS) while rocking vs. baseline during quiet sitting in rocking chair immediately prior to rocking; †p-

values corrected for multiple hypotheses using the Holm-Sidak method; ‡Due to discomfort, 3 children

did not use cervical paraspinal EMG sensors; *the confidence level was set to 95%; however, due to the

nature of the non-parametric signed rank test and algorithms used by the Prism 10 software used, 98.8%

and 97.9% CI were output for SCI and TD groups respectively for all muscles except for cervical

paraspinals where 99.2% CI were output for both groups; PS = Paraspinal; CI = confidence interval; IQR

= Interquartile range;

Table 12: Increase in muscle activation while rocking for TD children.

Muscle n

Baseline:

Median[IQR]

Rocking:

Median[IQR]

Median Increase

(CI) p-value†

Cervical PS‡8 6.86[5.37, 9.77] 17.0[11.3, 22.8] 8.09(4.80, 18.4)* 0.012

Thoracic PS 10 5.28[4.87, 6.40] 13.1[9.62, 21.5] 8.13(2.76, 20.4) 0.012

Lumbar PS 10 3.35[3.12, 3.90] 8.39[6.04, 13.6] 5.20(1.04, 10.1) 0.012

Pectoral 10 3.97[3.47, 5.33] 5.86[4.54, 11.7] 2.04(0.58, 9.15) 0.0137

Biceps 10 3.37[2.89, 4.19] 5.42[3.90, 7.96] 1.17(0.34, 5.22) 0.0136

Triceps 10 3.64[3.24, 4.81] 7.59[4.16, 8.81] 3.17(0.45, 5.79) 0.012

Rectus Abdominis 10 3.19[2.94, 3.50] 5.19[4.31, 7.43] 2.01(0.52, 5.84) 0.012

Oblique 10 4.13[3.33, 4.25] 6.76[5.13, 9.91] 2.80(0.99, 6.57) 0.012

Rectus Femoris 10 3.07[2.98, 3.62] 4.99[3.85, 6.83] 1.68(0.55, 6.67) 0.012

Hamstring 10 3.41[3.00, 3.87] 11.9[9.69, 18.9] 8.90(3.05, 19.3) 0.012

Tibialis 10 3.33[2.90, 4.96] 18.9[15.1, 29.0] 15.8(7.44, 25.8) 0.012

Gastrocnemius 10 2.81[2.52, 3.35] 5.75[4.38, 11.8] 2.79(1.46, 19.7) 0.012

All muscles tested showed a significant (p < .05) increase in muscle activity (as measured by mean of

EMG in µVRMS) while rocking vs. baseline during quiet sitting in rocking chair immediately prior to

rocking; ; †p-values corrected for multiple hypotheses using the Holm-Sidak method; ‡Due to discomfort,

2 children did not use cervical paraspinal EMG sensors; *the confidence level was set to 95%; however,

due to the nature of the non-parametric signed rank test and algorithms used by the Prism 10 software

used, 98.8% and 97.9% CI were output for SCI and TD groups respectively for all muscles except for

cervical paraspinals where 99.2% CI were output for both groups; PS = Paraspinal; CI = confidence

interval; IQR = Interquartile range

Muscle activation is measured by the increase from baseline of EMG amplitude

(VRMS). Due to discomfort, cervical paraspinal sensors were not used on several

participants in each group leading to a smaller n for this muscle, and a different

confidence interval. Median increase and confidence interval are shown for all.

4.3.5 Temporal Muscle Activation Patterns

Typical temporal muscle activation patterns during the rocking cycle for both SCI

and TD groups are shown in Figure 14.

Figure 14: Averaged muscle activation during rocking in children.; x-axis is one full

rocking cycle; y-axis is normalized to the amplitude of EMGRMS in the rocking cycle;

n.u. = normalized units

General trends indicate that many of the muscles exhibit similar activation

patterns between the groups but are often out of phase with each other. For instance,

cervical and thoracic paraspinals show very similar activation curves, but in the SCI

group, the peak activation occurs earlier in the rocking cycle. Other muscles have

maximum activations that occur close to the same point in the rocking cycle but exhibit

different shaped curves. For example, both the SCI and TD groups show maximum

activation of the rectus abdominis at the same point in the rocking cycle, but the SCI

group has a much broader activation peak than the TD group. Table 13 describes the

observed locations of peak muscle activation and the effects of muscle activation with

respect to the rocking cycle in SCI and TD groups.

Table 13: Observed actions of muscles during rocking in children with SCI.

Max Activation

Location

(% of cycle)

Muscle SCI TD Curve comparison Muscle Action

Cervical

Paraspinal

73%

(Mid FW)

93%

(Late FW)

Similar shape; SCI peak occurs earlier in cycle Stabilizes trunk, resists forward motion, and shifts weight back to

assist in rocking

Thoracic

Paraspinal

70%

(Mid FW)

89%

(Late FW)

Similar shape; SCI peak occurs earlier in cycle Stabilizes trunk, resists forward motion, and shifts weight back to

assist in rocking

Lumbar

Paraspinal

74%

(Mid FW)

93%

(Late FW)

SCI peak occurs earlier in cycle with secondary peak in BW phase Stabilizes trunk, resists forward motion, and shifts weight back to

assist in rocking

Pectoral 60%

(Late FW)

38%

(Late BW)

SCI peak occurs later in cycle Pull on arm rests as chair reverses direction at back of cycle

Biceps 50%

(Late BW)

42%

(Late BW)

Very similar curves with single peak occurring close to direction

reversal at back of cycle

Stabilizes and shifts weight forward at the end of back swing

Triceps 74%

(Mid FW)

85%

(Mid FW)

Max activation close to the same point, but TD has secondary peak

in late BW phase, and SCI does not.

Stabilizes and shifts weight backward at the end of front swing

Rectus

Abdominis

53%

(Mid BW)

36%

(Mid BW)

Max activation at similar location, but SCI has broader peak Stabilizes trunk and shifts weight forward in preparation for

reversal of direction

Oblique 25%

(Mid BW)

40%

(Mid FW)

Max activation occurs during backward motion, but SCI peak

occurs earlier in cycle

Stabilizes trunk and shifts weight forward in preparation for

reversal of direction

Rectus

Femoris

(Early

BW)

41%

(Late BW)

SCI peak occurs as chair reverses direction at front of cycle; TD

peak occurs just before back of cycle

Flexes hip

Hamstring 30%

(Late BW)

66%

(Early

FW)

TD and SCI both have secondary peaks that occur close to each

other’s primary peaks

Straightens hip as chair moves back

Tibialis

70%

(Mid FW)

71%

(Mid FW)

SCI and TD peaks both occur mid FW motion, but SCI peak is

broader

Lifts front of foot to avoid impeding forward motion

Gastroc 19%

(Mid BW)

32%

(Mid BW)

SCI peak that spans from before to after direction reversal at front

of cycle. TD slightly later.

Push on footrest during backward motion

FW = Forward motion phase of rocking cycle; BW = Backward motion phase of rocking cycle; Early = first third of motion in FW or BW direction; Mid = middle third of

motion in FW or BW direction; Late = last third of motion in FW or BW direction

Cluster analysis based on the timing of muscle activation divided the SCI group

into two clusters of seven and four subjects as shown in the dendrogram in Figure 15.

The average calculated distance from the seven and four-subject clusters to the TD group

was 0.78 and 1.16 respectively. Thus, the seven-subject cluster was labeled the “TD-

similar cluster”, and the four-subject cluster was labeled the “TD-dissimilar cluster”.

Plots of the clustering data for the trunk muscles are shown in Figure 16. This

figure illustrates that, for most trunk muscles, the temporal activation patterns in the TD-

similar cluster are closely comparable to those of the TD group, whereas the patterns of

the TD-dissimilar group are not.

Figure 15: Dendrogram showing clusters based on average normalized EMG

amplitude in each rocking cycle quartile for trunk muscles.

Figure 16: Average normalized EMG amplitude in trunk muscles during each quarter of the

rocking cycle for the TD cluster and two SCI clusters; *significant difference between

different cluster and TD cluster; †significant difference between similar cluster and TD

cluster

4.3.6 Muscle Activation Amplitude

Figure 17 shows a heatmap of muscle activation in each muscle as measured by

mean EMG amplitude (VRMS) during rocking. Each row represents one child, and each

Figure 17: Muscle activation (µVRMS) heatmaps for SCI group. ; Higher numbers and

darker colors represent higher activation; Muscles are on x axis, sorted by spinal cord level

of nerve roots. Y axis shows SATCo score, representing trunk control. SATCo =

Segmental Assessment of Trunk Control

column represents one muscle. Muscles are sorted from left to right in a cephalocaudal

direction. For children with SCI, the rows are sorted by SATCo score.

Results of Spearman correlation analysis of SATCo vs. muscle activation

amplitude for the SCI group are presented in Table 14. No significant correlation is

observed for cervical or thoracic paraspinals, but a significant positive correlation is

observed for rectus abdominis (r = 0.89, p = 0.003), oblique (r = 0.73, p = 0.03), and

lumbar paraspinals (r = 0.68, p = 0.05), indicating that participants with higher SATCo

scores show higher activation in muscles lower on the trunk than those with lower

SATCo scores.

Table 14: Activation amplitude during rocking vs. SATCo score

Participant SATCo

Cervical

Paraspinal

(µVRMS)

Thoracic

Paraspinals

(µVRMS)

Rectus

Abdominis

(µVRMS)

Oblique

(µVRMS)

Lumbar

Paraspinals

(µVRMS)

Mean TD 20 9.5 10.1 2.8 3.5 6.9

P15 19 6.7 13.6 15.4 17.1 15.0

P21 19 7.4 14.1 10.8 8.8 21.1

P217 18 NA†34.3 8.4 5.4 12.9

P204 15 6.9 7.7 3.6 1.5 2.0

P32 12 12.3 32.8 0.7 1.0 2.4

P187 12 NA†12.8 6.0 3.8 3.8

P23 11 27 -0.9 0.4 1.2 0.5

P34 11 4.6 4.8 0.5 0.5 0.6

P188 8 NA†28.8 0.7 1.8 6.1

Spearman’s r -0.18 0.30 0.89 0.73 0.68

P (two-tailed) 0.74 0.42 0.003* 0.03* 0.05*

*Significant difference (p<.05); †Cervical EMG sensors not used due to participant

discomfort; SATCo = Segmental assessment of trunk control, RMS = root mean square

4.4 Discussion

Of primary importance, the results confirmed our first hypothesis that trunk

muscle activation would increase above baseline during rocking. In fact, all muscles

tested in both TD and SCI groups showed significantly higher levels of activation during

rocking than at baseline. These results suggest that, in addition to activating trunk

muscles and being helpful for children with poor trunk control, rocking may also be a

useful activity for activation of leg muscles in children with SCI. Some children who can

activate leg muscles may still not be ambulatory due to weakness, poor balance, or

inability to manipulate an assistive device. Rocking may be an accessible form of

enjoyable, voluntary movement to use available trunk and leg muscles.

Our second hypothesis of differences between trunk muscle activation during

rocking by children with SCI and TD children was in part confirmed by cluster analysis

of temporal activation of muscles during the rocking cycle. This revealed two muscle

activation patterns, a pattern similar to TD and a pattern dissimilar to TD, providing

insights into the motor strategies employed by children with SCI during rocking. One

notable finding was the higher variability in muscle activation patterns in the SCI group,

and the presence of synchronous and counterbalance rocking strategies employed by the

children with SCI. These different rocking strategies are at least partially reflected in the

cluster analysis. For instance, three of the four subjects in the “TD-dissimilar” cluster

used the counterbalance strategy to rock, while none of those in the “TD-similar” cluster

used this strategy.

There is also a possible relationship between optimal fit in the rocking chair and

rocking styles. Both of the children with poor fit that could not be completely corrected

via padding or other adjustments were in the “TD-dissimilar” cluster, and all the children

in the dissimilar cluster (P216, P21, P187, and P217) needed extra padding on the

footrest, behind the back, or both due to issues with fit. In contrast, the “TD-similar”

cluster only contained one child who needed additional padding (P188 with 7.5 cm of

padding behind her back). Since the smallest and largest children were the most likely to

need extra measures to achieve good fit, it is also possible that both the issues with fit and

the differences in temporal muscle activation patterns were due to their size. Further

research will be needed to clarify the question of proper fit relating to meaningful muscle

activation.

A significant association was observed between trunk control, as measured by

SATCo score, and activation of muscles lower on the trunk. This matches expectations,

since higher SATCo scores reflect greater control and activation of trunk muscles in

lower thoraco-lumbar areas. It should be noted that the two subjects with AFM were

excluded from this analysis, since there is not a clear, level-dependent pattern of

impairment with this etiology.

Primary motivations for investigating the use of a rocking chair in this population

include accessibility and enjoyment. Importantly, there were no injuries during testing,

and the identified safety concern has been addressed by repositioning a cross brace so

that it no longer presents a potential to be a pinch point. It was found that all children

included in the study were able to make the chair rock, regardless of previous experience

or impairment. Feedback from children and parents during rocking also demonstrated that

the children enjoyed rocking in the chair and were motivated to use their muscles to make

it rock. These factors demonstrate that rocking gives children with SCI an enjoyable

opportunity to move independently, which may be rare outside use of a wheelchair for

mobility and thus particularly motivating for them.

4.4.1 Limitations

The results highlighted in this study should be interpreted considering identified

limitations. As noted previously, there were several sources of noise in the EMG data

used for this analysis. While all of these were addressed, it is important to acknowledge

that this does introduce some additional uncertainty regarding the accuracy of some of

the source data.

Additionally, the SATCo scores of the participants, which indicate their inherent

trunk control, ranged from 0 to 19. However, the distribution skewed towards higher

SATCo scores. Ideally, a more balanced distribution of SATCo scores would have been

preferred to ensure a broader representation of participants across the entire spectrum of

trunk control abilities.

While all children were able to rock, fit in the rocking chair may have had an

effect on muscle activation patterns, especially in cases where the balance of the rocking

chair was significantly shifted away from its normal neutral position, and this may have

effected the muscle activation patterns in some children, and influenced some results.

4.4.2 Future Work

In future research, enrolling children with a greater range of SATCo scores may

better quantify the relationship between trunk control and muscle activation patterns

during rocking and allow for exploration of additional strategies to activate the desired

muscles for rocking. For example, redesigning the footrest to attach to the seat of the

chair instead of the base could target the use of trunk muscles, as it would prevent use of

leg muscles initiating rocking by pushing the footrest. Additionally, asking children to

take their arms off the armrests could encourage the use of trunk muscles for weight

shifting stabilization during the forward and backward phases the rocking cycle.

Future work should also that the participants fit with the rocking chair may affect

the resulting muscle activation patterns. In addition to defining what constitutes

acceptable fit, parameters to fully characterize fit, such as chair balance point, footrest

height, and seat position should be defined and recorded. This data could then be

included in any analyses performed to learn how it affects rocking.

Investigating the long-term use of rocking chairs in the home would be valuable

to understand the different rocking strategies employed by children with SCI and the

progression of neuromuscular activation and capacity. Chairs could also be equipped

with sensors to monitor rocking performance, including parameters such as time spent

and amplitude of rocking. Sensors could also be installed to detect forces applied to the

rocking chair and provide valuable information about the muscles being utilized by the

child during rocking. Other possible uses of sensor data would be to adjust the difficulty

of rocking to provide extra resistance or help depending on the needs of the child, or to

provide feedback in the form of a game to encourage the use of target muscles.

4.5 Conclusions

Rocking in this prototype glider rocking chair activates the neuromuscular system

and is an accessible activity for children with varying degrees of trunk impairment due to

SCI. Additionally, the study provides the first examination of the timing and amplitude

of muscle activation during rocking, and the differences between the TD and SCI groups.

These results, together with the observation that the rocking chair prototype is safe and

enjoyable for use by children in this population, highlight the potential of rocking chair-

based interventions as an effective approach for trunk muscle activation and a promising

component of extending ABT in children with SCI beyond the clinic to the home and

community.

CHAPTER 5

MUSCLE ACTIVITY PREDICTION FROM EMBEDDED SENSORS

5.1 Introduction

In children with SCI, lack of trunk control due to neuromuscular paralysis/paresis

has been shown to increase the risk of secondary conditions such as scoliosis and

compromised lung function [3, 13, 17, 27, 28]. Recent research has shown that activity-

based therapy (ABT) is effective at improving trunk control in this population [33]. In

order to address the need for children to continue to activate trunk muscles in the home, a

rocking chair designed for children with SCI, and been designed, built and tested as

described in chapter 3 [59].

To validate that rocking in this rocking chair activates muscles, TD children and

children with SCI rocked in the chair while muscle activity was measured using EMG.

As described in Chapter 4, comparisons were drawn between muscle activation patterns

in the two cohorts, and it was established that muscle activity during rocking is

significantly higher than at baseline for muscles in arms, legs, and trunk.

The current chapter will build on this previous work by incorporating various

sensor technologies into the chair to provide feedback to therapists about user activity

while rocking in the rocking chair. For example, sensors in the chair would enable

therapists to track basic use factors such as rocking duration and amplitude.

Appropriately located force sensors would be used to monitor specific activity of the

limbs and trunk and therefore may eliminate the need to use EMG which, while an

accepted standard for measuring muscle activity, is expensive and requires training.

Information about the which muscles are employed by the child to make the chair rock,

as well as insight into the ways muscle use patterns change over time would provide

useful information for therapists interesting in longitudinal tracking of improvements

provided by muscle activation activities. By adding multiple sensors in strategic

locations to measure forces applied to the rocking chair and body movements initiated by

the child, a variety of rocking activity data can be collected.

Initially, specific aim 4 focused on using sensor data to develop methods to

predict SATCo score using multivariate linear regression analysis. As the project

progressed, this was found to be problematic since each child only provided a single data

point with respect to SATCo score and not all SATCo scores were represented in the

dataset. Additionally, correlations between sensor data and trunk muscle activation were

too complex for standard linear regression techniques to be effective. To address these

difficulties, machine learning techniques were implemented, and the focus was shifted

from predicting SATCo score, to predicting muscle activation during rocking.

5.1.1 Study Rationale

In this chapter, data recorded from sensors embedded in the prototype rocking

chair will be used to quantify chair and participant motion, as well as forces applied by

users to train machine learning models in an attempt to infer subject muscle activations

during rocking. Machine learning techniques used in this research will include both

established regression learning and neural network model techniques.

5.1.2 Machine Learning Application to Chair Sensors

Biomechanical modeling using inverse dynamics calculated from 3-dimensional

motion capture and force-plate data has been used to assess loads on internal anatomical

structures including joints and muscles [64]. In the context of rocking in the home,

however, biomechanical modeling cannot easily be used to quantify muscle activation

forces. This challenge stems from multiple factors, including the requirement for costly

motion capture systems, which are not available in the home or even in most clinical

settings, and the need for accurate measurements of multiple reaction forces on the

rocking chair. Additionally, factors such as the need for custom biomechanical modeling

for each child, and complicated setup procedures for each rocking session complicates

this approach.

For similar situations where motion capture and/or force plate data cannot be

obtained, machine learning methods have been used to aid in biomechanical analyses

[64]. Examples include estimation of ground reaction forces based on body-worn

accelerometer data [65, 66] and estimation of voluntary elbow torque based on EMG and

kinematic data [67]. In addition to analysis of biomechanical problems, machine learning

techniques have been used to draw inferences based on data from sensors embedded in

equipment used by patients. An example of this is the use of Logistic Regression and

Feed Forward Neural Networks to categorize the orientation of patients in bed based on

data from load cells positioned under the legs of the bed. This system could help to

reduce the incidence of pressure sores by alerting caregivers when a patient’s position

has not changed after a prescribed period of time [68].

5.1.3 Regression Learning

Regression learning is a supervised learning technique that is used to find the

correlation between independent input variables and a dependent output variable. In this

work, ensemble learning, which utilizes multiple regression models to learn relationships

between a set of features and a response, will be used. These models are then combined

to produce a stronger learner with improved generalization performance. The base

learner that was selected for this work is the decision tree, and boosting was chosen as

the approach to constructing the composite learner. Boosting is an incremental learning

process for constructing the composite learner, in which new learners compensate the

error from previous learners. [69]

5.1.4 Neural Networks

Briefly, a neural network is a common machine learning model built from layers

of interconnected artificial neurons. A typical neural network consists of an input layer,

an output layer, and one or more hidden layers. Neurons in each layer are interconnected

nodes with neurons in neighboring layers, and weights are assigned to these connections

[70]. Training of a neural network involves adjusting these weights to minimize the error

between the predictions and the actual targets. Learning algorithms, such as

backpropagation, where errors are propagated back through the network from the output

layer to reduce overall error, are used to adjust weights. Regularization techniques can

also be applied to prevent overfitting, which causes the model to perform well on training

data, but poorly on new data. Regularization can help maintain smaller weights and

produces models that are better at generalizing to new data [71].

Using these techniques, models will be trained to infer muscle activation in 12

individual muscles in the user’s arms, legs, and trunk. Finally, the performance of the

model will be compared against individual sets of activity data to evaluate efficiency of

predicting activation of specific muscles during rocking by SCI patients evaluated under

IRB approval.

5.2 Materials and Methods

5.2.1 Participants

Participants for this analysis were the same as detailed in sections 4.2.1, 4.2.2, and

4.3.1, and were enrolled under University of Louisville IRB #17.0725. Two cohorts were

enrolled: children with SCI and TD children. EMG data collected from these participants

(as detailed in section 3.2.4) was also used in this analysis.

5.2.2 Rocking Chair Instrumentation

As detailed in Chapter 3, the prototype rocking chair was designed to include

embedded sensors which could capture forces applied to different parts of the rocking

chair, and movement of the seat and the subject’s trunk during rocking. Sensors used,

sensor locations, and data captured are shown in Table 13.

Single zone force sensitive resistors were installed on the arm rests (Walfront

SF15-150 Force Sensitive Resistor, sensitivity range 0.1 – 100N), and in the seat of the

chair (Interlink Electronics FSR® Model 406, sensitivity range 0.1 – 10N) to track forces

applied during rocking. In order to increase the sensitivity of the FSRs, actuators were

designed, and 3D printed to concentrate the applied force. For instance, the actuator for

the arm rest FSR (Figure 18) sat between the armrest cushion and the FSR and helped to

ensure that force applied to any part of the cushion would trigger a similar response from

the FSR. Similarly, the seat FSRs were mounted on a flexible 3D printed mat and 3D

printed actuator positioned over them to enhance their sensitivity to forces applied to the

seat padding overlaying them.

Table 15: Sensors used to collect data for training machine learning models

Sensor Name Sensor Type Measurement Taken

Footrest Four half-bridge load cells configured as

eight active element Wheatstone bridge

Force applied by feet to top of footrest

Right arm rest FSR under pad on right armrest Force applied to top of pad

Left arm rest FSR under pad on armrest Force applied to top of pad

Trunk Strap Optical Time of Flight sensor Elongation of strap around trunk as tension

is applied.

Position Rotary potentiometer driven by motion of

linkage arms

Seat position in rocking cycle

Seat Four FSRs in square array Force applied to each seat FSR

CoP Calculated from seat FSRs Forward/back movement of pressure

applied to the seat. Average of front FSRs

minus the average of back FSRs.

Accelerometer Accelerometer Acceleration parallel to swing of the

rocking chair

FSR = Force Sensing Resistor; CoP = Center of Pressure

The footrest was constructed of two sheets of Baltic Birch plywood, with four

recesses cut in the lower piece and load cells (SMAKN® Half Bridge Body Load Cell

Electronic Scale Weighing Sensor 50Kg) set into them. The top sheet of plywood was

then placed on top of these load cells and attached with screws which passed through

holes drilled in the lower sheet of plywood, which allowed the upper sheet to move

vertically, but constrained horizontal motion. This allowed any load placed on the footrest

to actuate the load cells. Load cells were connected in an eight active element Wheatstone

bridge configuration, and changes in voltage were read to measure the applied loads.

Figure 18: Armrest FSR with 3D Printed actuator

As previously detailed in section 3.5.3, a position sensor was constructed of a

rotary potentiometer (TT Electronics, P160KNP) installed on the base of the chair and

coupled to one of the chair’s rocker arms via spur gears. This setup can track the position

of the chair throughout its entire range of motion. The potentiometer is configured as a

voltage divider (5V input, 10k ohm fixed resistance, 10k ohm variable resistance), and

the voltage read from the center pin corresponds to the position of the chair.

To track trunk motion, a novel sensor was utilized. This sensor has been

described in detail in Lin et al [72], but briefly, consists of stretchable fibers with a

urethane core and a silicone cladding sandwiched between two pieces of elastic fabric

with the ends routed, via a 3d printed mounting block, to a 5 x 5 mm mirror oriented to

couple the fiber ends to the ports of a miniature time-of-flight (TOF) sensor for light

detection and ranging (LIDAR) (Pololu VL53L0X ToF sensor, Pololu Robotics and

Electronics). This

Figure 19: Optical stretch sensor on belt mounted on rocking chair

arrangement allows the TOF chip to detect elongation of the optical fiber as the fabric

is stretched. This sensor was attached to a belt (AliMed® Pediatric Walker Belt) using

hook-and-loop fasteners at each end, so that the elastic portion of the fabric would

elongate when tension was applied to the belt. This belt was attached to the seat back of

the chair as shown in Figure 19 and fastened around the trunk of the subject during

rocking so that trunk movement would stretch the optical fiber, producing a signal from

the TOF chip which was recorded. It should be noted that, as this sensor became

available after data had been collected from the first seven children with SCI. Data from

the strap is available for the last four children with SCI who were tested, and for all TD

children.

All sensors were connected to a custom PCB, which was designed to plug into

IO ports A and B on an NI myRIO 1900-(National Instruments, Austin, Texas) as shown

in Figure 20. Schematics are included in Appendix C. A load cell amplifier (SparkFun

Qwiic Scale NAU7802, Sparkfun Electronics, Niwot CO) was mounted on the PCB to

acquire data from the load cells in the footrest. I2C data from the load cell amplifier and

the TOF sensor were acquired by an Arduino Nano (Arduino.cc) mounted on the custom

PCB and relayed to the myRIO-1900 where all data was saved to file.

Figure 20: Custom PCB for data collection

5.2.3 Data Collection

sEMG data collected as previously detailed in section 3.2.4 was used to provide

the target data for training and testing the machine learning models. Data from sensors

embedded in the rocking chair was acquired at 40 Hz, and a pulse signal was recorded at

the beginning and end of each data capture session to facilitate synchronization between

the 2000 Hz sEMG data and the sensor data.

5.2.4 Regression Model Development

5.2.4.1 Data Processing and Feature Extraction

Four FSRs in the seat of the chair positioned in a square array were used to

generate an approximation of the center of pressure applied to the seat. This was

calculated by taking the average of the front FSRs minus the average of the back FSRs.

This gives a measure of fore-aft changes in pressure as the child rocks and can be used

detect weight shifts due to trunk movement. This data was treated as a separate sensor,

and features were extracted from it as well as from the individual seat sensors.

To synchronize the sEMG data with the sensor data, it was necessary to

increase the data rate of the sensor data to match the 2000 Hz rate of the sEMG data.

This was done with a custom MATLAB script (Appendix A) which identified the

beginning and end of each data collection in both data files using the synchronization

pulse, and the sensor data was interpolated to match the length of the sEMG data.

5.2.4.2 Artifact Removal and Rocking Cycle Creation

As detailed in section 3.2 artifacts were removed from sEMG data, and data was

broken into individual rocking cycles. Each rocking cycle was standardized to 1000 data

points in length, and sensor data was normalized by to the peak-to-peak amplitude of the

corresponding cycle using equation 1, similar to the method used previously in prototype

EMG cycle creation, and described in Elamary et al. [73, 74].

Features were extracted for each rocking cycle for use as inputs in training of the

regression model. The features listed in Table 14 were calculated based on data from

each sensor using a custom MATLAB script (Appendix A). Features were calculated as

follows. Points 1-10: The 1000 data points for the given cycle were coarse-grained down

to 10 points (P1 – P10) by stepping through the data 100 points at a time and taking the

average of those points. Thus, the first point (P1) represents the average of the first 100

points in the cycle, and P2 represents the average of points 101-200 and so forth. Area

under the curve was calculated numerically simply by summing the value of all 1000 data

points for the given cycle. Maximum and minimum values were simply the largest and

smallest data point in the given cycle. Average was calculated by taking the mean of all

data points in the cycle. Peak-to-peak amplitude was the difference between the

maximum and minimum values in the given cycle. Maximum and minimum locations

were simply the index (from 1 to 1000) of the maximum and minimum data points in the

cycle. RMS was calculated by taking the square root the mean of the squares of all data

points in the cycle. Maximum Sum and maximum sum location were calculated by

finding the value of the largest sum of 300 contiguous points, and the index of its center

point. Frequency points 1-10 (F1 – F10) were calculated by taking the power spectrum

between 0 and 6 Hz, split into 10 bins, with the value of each bin representing one

feature. Centroid of the power spectrum with respect to frequency was calculated by

dividing the sum of the ten frequency points by the number of points (10).

Table 16: Features extracted from sensor data to use when training machine learning

models

Feature Description

Points 1-10 y-value of data points in given cycle, course-grained to 10 points, giving 10

features per sensor

AUC Area under the curve for the given cycle

Max_Value Maximum y-value in given cycle

Min_Value Minimum y-value in given cycle

Average Mean y-value of all data points in the given cycle

P2P_Amplitude Peak to Peak Amplitude: Max cycle y-value – Min cycle y-value

MaxLoc Location on the x-axis of the maximum value in the given cycle

MinLoc Location on the x-axis of the minimum value in the given cycle

RMS Root mean square value of all data points in the cycle

MaxSum Value of largest sum of 300 contiguous points

MaxLoc Center of max 30% of data (largest sum of 300 points)

Frequency 1-10 Power spectrum between 0 and 6 Hz split into 10 bins, giving 10 features per

sensor

Centroid Location of centroid of power spectrum with respect to frequency

5.2.5 Regression Models

Due to challenges encountered when normalizing the EMG readings in SCI

patients, where no maximum exertion is possible, selecting the normalization method

that best corresponds with the amount of force exerted by the muscles was difficult.

Several normalization techniques were trialed, including normalization by subtracting the

baseline EMG reading before rocking started, using the peak-to-peak amplitude of the

EMG readings during rocking, and normalizing by dividing each reading by the

maximum reading recorded during rocking. Ultimately, training targets were based on

EMG readings which were normalized by subtracting the baseline reading from each data

point.

For training, data was segmented into a training data set and test datasets. First,

data from each participant was segmented in half, and the first half of the data from all

participants was assembled into a training dataset. The second half of the data for each

participant was used to test the trained regression model.

Two regression models were developed. The first was developed using multiple

linear regression in Prism 10 (GraphPad Software, San Diego, CA, USA). First,

predictors were standardized by scaling to have a mean of zero, and a standard deviation

of 1. Variance inflation factors (VIF) were calculated and were used along with a

correlation matrix to manually remove predictors with a high degree of

multicollinearity. Once all predictors had a VIF < 10, p-values of the remaining

predictors were used to manually eliminate predictors which contributed little to the

model. Multiple linear regression was then performed, and β values were produced for

each predictor. This

procedure was repeated for each muscle. A custom MATLAB script was then used to test

the models using testing data from each child.

Another regression model was also created, using the MATLAB regression

learner. To evaluate for the optimal training method, sample datasets were trained all

available methods including linear regression, regression trees, support vector machines,

Gaussian process regression, kernel approximation regression, and ensembles of trees.

Models were initially screened by comparing mean squared error (MSE) for sample

datasets. A custom MATLAB script (Appendix A) was used to train the model using

the ensembles of trees method, with the boosted trees option.

5.2.6 Neural Network Model Development

A nonlinear input-output time-delay neural network was created and trained in

MATLAB, using sensor data as the predictors and EMG data as the targets. As with the

data for the regression model, to standardize the length of a rocking cycle across subjects

and cycles, and to ensure that time delayed inputs would always be representative of the

same point in the rocking cycle relative to the target, data were interpolated to 1000 data

points per rocking cycle. To optimize the network’s parameters, several models were

trained, altering input delays and the number of layers.

EMG and sensor data from each subject was split into training, testing, and

validation datasets, with training data comprising the first 60% of the data, and testing

and validation data comprising the 20% immediately following the training data, and the

final 20% respectively. Since muscle activation, which provided the target data for

training the neural network in the form of EMG data, was driving the sensor output, the

sensor data was offset from the EMG data, so that the inputs used to predict a specific

target value were taken from the rocking cycle immediately following the target as well

as the rocking cycle previous to the target time point.

The network was trained using Bayesian Regularization backpropagation, a

network training function that uses Levenberg-Marquardt optimization to update weight

and bias values. By minimizing a combination of squared errors and weights, it penalizes

large weights and produces a network that generalizes well [74, 75].

The network was set to 8 hidden layers, 11 inputs based on the 11 sensor values,

and 12 outputs for muscle activation targets as shown in Figure 21. Input delays were

Figure 21: Neural Network diagram with 11 inputs, 8 hidden layers, and 12 outputs

between 0 and 2000 with a step-size of 100, giving an array of 21 values for each input (1

at the same time point as the target, 10 from the previous cycle, and 10 from the

subsequent cycle). Maximum number of training epochs was set to 100, with a stop

condition of 10 consecutive validation failures. Separate models were trained for each

child, and trained models were tested using the prepared testing datasets.

5.2.7 Model Evaluation

The multiple linear regression model, ensemble regression model, and the NN

model performance were tested using the test dataset previously created. Several

measures of model performance were calculated based on the predictions produced by

the entire test dataset. These include R2, Mean Absolute Percentage Error (MAPE), Mean

Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE),

and Sum of Squares (SS) [76], which were calculated for each model and compiled into a

table for comparison.

To investigate the utility of the models in predicting muscle activation patterns in

individual children during rocking, the test data for each child in the SCI group was used

to predict muscle activation with each model. Correlation of the means of predictions

with targets for each muscle was calculated to evaluate how well the activation patterns

identified by the predictions matched the observed muscle activation. Many of the

datasets failed normality tests, so Spearman correlation analysis was selected, and results

were compiled into a table.

5.2.8 Evaluation of Sensor Data Correlations

Features were evaluated for correlation with muscle activity using the Minimum

Redundancy Maximum Relevance (MRMR) algorithm in the MATLAB regression

learner. For each muscle, the feature that was rated highest was tabulated and results

were compiled into a table showing the sensors that were most correlated with trunk,

arm, or leg muscle activity.

5.3 Results

5.3.1 Model Comparison

Table 17: Comparison of error for muscle activity prediction models

R Squared MAPE MSE

Multiple

Linear

Regression

Decision

Tree

Regression

Neural

Network

Multiple

Linear

Regression

Decision

Tree

Regression

Neural

Network

Multiple

Linear

Regression

Decision

Tree

Regression

Neural

Network

Cervical PS 0.000 0.226 0.402 229% 127% 59% 179.7 70.2 352.3

Thoracic PS 0.115 0.506 0.466 724% 309% 56% 583.4 79.3 169.5

Lumbar PS 0.006 0.411 0.441 738% 265% 72% 454.6 66.6 112.3

Pectorals 0.085 0.336 0.500 1996% 1459% 45% 502.8 64.2 106.2

Biceps 0.109 0.464 0.460 2641% 1394% 113% 1973.6 672.2 910.6

Triceps 0.246 0.434 0.444 751% 772% 87% 515.6 135.1 336.8

Rectus Abdominis 0.140 0.440 0.758 1359% 446% 58% 1218.6 84.1 67.1

Oblique 0.008 0.162 0.591 506% 206% 42% 115.8 38.4 28.9

Rectus Femoris 0.008 0.299 0.472 3335% 558% 61% 963.1 107.7 163.9

Hamstring 0.033 0.429 0.578 4082% 441% 56% 2908.2 73.0 96.5

Tibialis Anterior 0.007 0.261 0.584 7811% 1428% 66% 2200.5 234.8 242.8

Gastrocnemius 0.004 0.572 0.420 13809% 964% 81% 4176.1 90.4 189.4

R2 = R Squared; MAPE = Mean Absolute Percent Error; MSE = Mean Squared Error; PS = Paraspinals

Table 15 compares several measures of error in each of the models. R2 describes

the proportion of the variance in observed activation in the given muscle that is described

by each of the models. R2 values for successful regressions typically range between 0 and

1, with higher values indicating a more successful regression. MAPE quantifies error in

terms of percentages, making it suitable for comparing models with target data that has

been scaled differently, or when models have been tested on different data, as in the case

of the neural network and the regression models. Mean squared error tends to be more

sensitive to outliers than other measures of error and is the metric that is used by

MATLAB to evaluate model performance during training [76].

R2 and MAPE were the primary metrics used to compare models, due to their

ability to compare models with outputs that are scaled differently from each other. By

both these measures, the neural network outperformed both regression models, with the

decision tree regression receiving a higher R2 value than the neural network for only three

muscles (thoracic paraspinal, biceps, and gastrocnemius) and multiple linear regression

scoring worst in both measures for all muscles except the triceps, where it scored very

slightly (751% vs 775%) better than decision tree regression.

5.3.2 Muscle Activation Pattern Prediction

Spearman correlation analysis of muscle activation predictions is shown in Table

16. Multiple Linear Regression predictions were significantly correlated (p < .05) with

targets for four of the eleven children with SCI, while Decision Tree Regression

predictions were significantly correlated with targets for five children. Neural Network

predictions were significantly correlated with targets for all children.

Table 18: Spearman Correlation Analysis of Predictions for SCI Group

Participant

ID Multiple Linear Regression Decision Tree Regression Neural Network

P15 -0.161 CI (-0.683, 0.470) p = 0.619 0.371 CI (-0.276, 0.786) p = 0.237 0.937 CI (0.778, 0.983) p = <0.0001*

P21 0.112 CI (-0.508, 0.656) p = 0.733 0.727 CI (0.245, 0.921) p = 0.01* 0.958 CI (0.848, 0.989) p = <0.0001*

P23 0.587 CI (0.001, 0.873) p = 0.049*0.35 CI (-0.298, 0.777) p = 0.266 0.979 CI (0.947, 0.996) p = <0.0001*

P32 0.636 CI (-0.020, 0.868) p = 0.03*0.573 CI (-0.020 to 0.868) p = 0.055 0.993 CI (0.973, 0.998) p = <0.0001*

P34 0.294 CI (-0.622, 0.549) p = 0.354 -0.056 CI (-0.622, 0.549) p = 0.869 0.965 CI (0.872, 0.991) p = <0.0001*

P215 -0.133 CI (0.61, 0.968) p = 0.683 0.881 CI (0.61, 0.968) p = 0.0003* 0.895 CI (0.65, 0.972) p = <0.0001*

P187 0.045 CI (-0.583, 0.640) p = 0.903 0.445 CI (-0.230, 0.831) p = 0.173 0.748 CI (0.153, 0.919) p = 0.007*

P188 0.882 CI (0.585, 0.970) p = 0.0007*0.673 CI (0.102, 0.910) p = 0.028* 0.993 CI (0.963, 0.998) p = <0.0001*

P216 0.273 CI (-0.374, 0.741) p = 0.391 0.517 CI (-0.099, 0.847) p = 0.089 0.986 CI (0.947, 0.996) p = <0.0001*

P204 0.294 CI (-0.354, 0.751) p = 0.355 0.636 CI (0.079, 0.890) p = 0.03* 0.979 CI (0.973, 0.998) p = 0.001*

P217 0.764 CI (0.284 to 0.938) p = 0.0086*0.682 CI (0.119, 0.913) p = 0.025* 0.832 CI (0.324, 0.943) p = <0.0001*

Spearman r (95% Confidence interval) p-value of correlation between the mean muscle activation

predictions and targets for each child in the SCI group; *Significant (P < .05) correlation between

predictions and targets

Figure 22 compares predictions from each of the three models for one of the

participants with SCI (P188). For this example, which showed significant correlation for

all models, one can see that, although the distribution of predictions for a given muscle is

often different than that of the targets, the general muscle activation pattern is similar.

Thus, a therapist looking at these predictions would be able to get an idea of which

muscles were primarily being activated during rocking, even if the exact magnitude of

activation was uncertain.

Figure 22: Example of muscle activation pattern predictions vs. target muscle

activations for one participant with SCI (P188). Note that y scale for Neural Network

predictions is different than for regression models due to differences in calculating

targets. Activation is expressed in µVRMS.

5.3.3 Sensor Correlation with Muscle Activity

MRMR analysis results are shown in Table 17. MRMR analysis revealed that the

strap data produced more highly relevant features (29) than any other sensor. However,

if all seat sensors are considered together, it surpasses the strap-based features with 39

highly relevant features. Strap features were highly relevant only for trunk and arm

muscles, with no highly relevant features related to leg muscle activity. Other sensors

showed a similar pattern of being more relevant to muscles they interacted more closely

with. The footrest produced features that were more relevant to leg muscles (6 for legs, 3

for trunk, 0 for arms). Seat sensors produced highly relevant features for muscles in all

domains with relatively balanced correlation to both upper and lower body muscles (19

for legs, 21 for upper body).

Table 19: Minimum Redundancy Maximum Relevance Analysis of correlation between

features derived from each sensor and muscle activity in trunk, arms, and legs

Sensor

Trunk

Extensors

Trunk

Flexors Arms

Legs

Extensors

Legs

Flexors Total

Strap 15 7 7 0 0 29

Left Arm 0 1 5 2 1 9

Right Arm 2 2 2 1 0 7

Seat 0 1 1 1 1 1 5

Seat 1 0 0 2 1 1 4

Seat 2 2 1 2 2 2 9

Seat 3 3 2 2 3 4 14

CoP 0 3 0 3 1 7

Footrest 2 1 0 3 3 9

Accelerometer 3 2 4 4 6 19

Position 2 0 4 0 1 7

Count of how many features based on each sensor were the most correlated

with different muscle groups; CoP = Center of Pressure

5.4 Discussion

5.4.1 Model Utility

There are two main metrics for evaluating whether the developed models are

useful to therapists. First, while exact determination of muscle activation at a given point

in time is not necessary, knowledge of the average activation of a given muscle relative

to other muscles during a rocking session would be useful in helping therapists determine

muscle use patterns, and to track changes in muscle use patterns over time. Correlation

analysis shows that the neural network performed better than either regression analysis,

with predictions significantly correlated to targets for all children.

Although correlation does not guarantee that the magnitude of each prediction is

close to the magnitude of the targets, Spearman correlation does show that the rank of the

corresponding predictions and targets are similar, and statistically significant correlation

shows that when ranked from most to least active, the order of predictions is similar to (if

not precisely the same as) the targets.

Second, models should at minimum detect whether a specific muscle is “turned

on” during rocking or whether it is inactive. With respect to this metric, the neural

network model performed well, accurately predicting very low activation (arbitrarily set

to < 5 µVRMS for purposes of this analysis) in 34 out of 41 cases, and only incorrectly

predicting low activation in two cases. The regression models, on the other hand, tended

to overestimate leg muscle activity for most children, while also underestimating activity

in those with significant leg muscle activation.

5.4.2 Sensor Relevance

In addition to the predictions produced by the models, the MRMR analysis

provided valuable information about the contribution of each sensor to predictions.

Importantly, sensor data was found to be most relevant to activity in muscles that they

interacted more closely with. For instance, force sensors in the footrest produced data that

was relevant to leg muscle activation, while sensors near the trunk (strap, armrest)

produced data more correlated with trunk muscle activation. This helps to reinforce the

conclusion that the sensor data contains information about muscle activation, and that the

machine learning models are using this to create their predictions, rather than simply

overfitting based on unrelated data. Knowledge of how sensors contribute to the models

for each muscle can also guide future sensor placement and feature selection decisions.

5.4.3 Limitations

The predictions made by these models are based on test data gathered in the same

session as the training data so, while the data used for testing had not been specifically

seen by the model during training, it was relatively similar in terms sensor data and EMG

target data. Thus, it is likely that at least some of the results are based on the model

learning to recognize the sensor data unique to a particular child and making predictions

consistent with muscle activation typical for that child. Until data can be collected across

several sessions in which there are differences in muscle activation patterns and in sensor

data collected it will be difficult to tell how well the model is able to detect changes in

muscle activation across time and across rocking sessions. Similarly, we cannot at this

point quantify how well this model would generalize to other subjects who had not had

the model trained on their data.

5.4.4 Future Work

There is need to collect additional data to assess the generalizability of the current

models across subjects and over time. These expanded datasets can be used to test the

accuracy of current models when applied to new data, and to facilitate their extending

their predictive capacity to a broader range of situations.

Current models produce data that is potentially useful to therapists, but a more in-

depth analysis to determine the level of accuracy and reliability required for predictive

models to be of practical use is needed. This will entail a systematic investigation to

ascertain the specific thresholds and criteria necessary to render predictions clinically

meaningful.

Future investigations should also expand on the work done on correlations of

different sensors with muscle activation to explore alternative types and locations for

sensor placement, such as the back of the seat, to enhance the accuracy and precision of

predictions.

To translate these findings for clinical use, it will be necessary to continue to

work with therapists to define the best measures to machine learning algorithms toward,

and to develop tools to provide the collected data to therapists in ways that are,

convenient, timely and useful.

5.5 Conclusion

This study demonstrates the potential of instrumented rocking chairs for

predicting muscle activation in pediatric SCI patients. Current models demonstrate that

machine learning techniques can be used to extract information about muscle activation

from force and motion sensors embedded in a rocking chair. While questions remain

about how well these results can scale across time and subjects, further research is

warranted to refine and validate these models for a broader population of pediatric SCI

patients. By exploring these avenues, future research can advance the efficacy and

applicability of predictive models in clinical settings.

CHAPTER 6

CONCLUSIONS AND FUTURE

DIRECTIONS

6.1 Research Objectives

Affectionately known as Rockin’Rehab, this project was structured to accomplish

four major objectives focused on the needs of children with spinal cord injury through the

development of a simple yet novel mechanism to activate trunk muscles.

The first project aim was to design and build a rocking chair that could be used by

children with SCI. The hope was that rocking would prove to be an accessible activity

that these children could do on their own regardless of their level of impairment, and that

in conjunction with other components of an ABT program, would augment the gains in

trunk control made through participation in locomotor training. Of primary importance

was to design and fabricate a rocking chair that would accomplish these goals in a safe,

enjoyable way.

The second project aim was to investigate how rocking in the rocking chair

affected muscle activation in children with SCI as compared to TD children. The

primary research question was whether muscles of interest (primarily trunk muscles, but

also muscles in arms and legs) would be activated during rocking. Another goal was to

describe both temporal muscle activation patterns and which muscles typically would be

used to drive rocking.

The third and fourth aims of the project were both centered around the use of

various sensors to instrument (sensorize) the rocking chair, which could be used to gather

100

data about the rocking experience that would be of use to therapists in the ABT program.

This would involve design and implementation of the sensor placement and data

collection systems, and the development of algorithms to detect muscle activity based on

the collected data.

6.2 Key Findings and Implications

The most important outcome of this study was the discovery that trunk muscles,

as well as muscles in the arms and legs, are activated by rocking. This significant finding

helped to achieve the second aim, which confirmed that rocking is a promising activity

for use in conjunction with a program of ABT.

The successful build of the rocking chair, and validation that it provides a safe,

enjoyable, and accessible activity for children with SCI was also an important outcome of

this research. The use of the QFD design method helped to ensure that the needs of both

the therapists and users were met and provides a solid reference point as further

developments and improvements to the rocking chair are implemented. The rocking chair

prototype was also found to be accessible and enjoyable to children with varying levels of

impairment, which helps to add confidence that children will use it.

The analysis of data collected by commercially available sensors embedded at

strategic locations in the chair to detect muscle activity was a novel idea and was the

most exploratory of the major objectives of this research. Although more work remains

to be done to fully develop these techniques for clinical use, the current results are

promising, and this technique may be applicable to other rehabilitation contexts.

101

6.3 Challenges and Limitations

Some limitations related to the design of the rocking chair were discovered during

rocking sessions with participants. Most of these were related to the ability of the rocking

chair to adjust and accommodate children of different sizes. While the chair enabled

children over a wide range of size to rock, some of the adjustments necessary to make

this possible disrupted other factors such as balance. This may in turn have influenced

muscle use patterns during rocking. Rethinking the methods of adjusting the chair in

ways that mitigate these interactions may be possible. For instance, instead of moving the

child in the seat to fit the distance to the footrest to the user’s leg length, the footrest

could be made to move backward and forward as well as up and down. Similarly,

physically adjusting the depth of the seat instead of changing its effective depth by

padding the backrest could address problems with fit, without changing the balance of

the seat.

Additionally, while fit was considered in setting up the rocking chair for the user,

it was not carefully characterized to investigate for its effects on muscle activation

patterns and its potential role as a confounding factor in the analyses performed. Data

that was collected about fit suggests that some factors (particularly balance) may have

had an effect on temporal muscle activation patterns, but without more precise data and a

larger sample size it is difficult to verify and quantify these effects.

Sensors used for sensing forces applied to the rocking chair were chosen for their

form factor which integrated easily into the rocking chair without interfering with its

operation. Some sensors, particularly the FSRs used in the seat and armrests, can suffer

from issues with hysteresis and drift. To some extent these issues are mitigated by the use

102

of standardization techniques used in machine learning and regression, where the data is

scaled to a mean of zero and an SD of 1. The time-dependent nature of hysteresis and

drift may, however, introduce noise into the data that isn’t completely removed by

standardization techniques, and complicate the attempts to glean meaningful information

about muscle activation from it.

Another challenge to training the machine learning models to predict muscle

activation was that all training data came from one relatively short rocking session, with

no major changes in rocking style or muscle use patterns over the course of the session.

This made it difficult to confirm that the models that were developed would generalize to

subjects whose data was not in the training dataset, or even to data taken from the same

subjects at subsequent rocking sessions. Until more data can be collected and models can

be refined, their reliability and utility will remain in question.

Finally, the temporal muscle activation patterns found for children with SCI are,

in some cases, likely to be an average of several different activation patterns rather than a

true representation of typical pattern for children with SCI. This may be due to children

with SCI possessing a wide variety of muscle activation patterns due to the need to

compensate for individually unique impairments. It is likely that instead of attempting to

define one typical pattern from children with SCI, the emphasis should be on defining the

patterns that are typical for TD children, and to study the ways that different impairments

alter muscle activation patterns to differ from what is typically seen.

6.4 Future Directions

Over the course of the project, several observations and suggestions for

improvements and additional features for the rocking chair were identified. Suggested

103

changes include items such as a larger footrest with straps to secure the feet, two or more

rocking chair models of different sizes, wheels to make transport of the chair easier, an

interlock to prevent chair rocking during transport, and other practical suggestions. These

adaptations should be examined to determine the need they are intended to meet, and

evaluated with for how well they meet the identified need in the same QFD manner from

the original chair build. Any features that fill an identified need should be implemented in

future versions of the rocking chair, or an alternative feature should be developed to meet

that need. Adult SCI patients may also benefit from the natural activity of rocking, and

additionally design parameters should be considered when developing for this population.

One goal of this project has been to move toward making rocking chairs available

for use in children’s homes, a hint towards the potential for commercialization of the

chair design. But this is also a reasonable proposition since it would allow patients to

further the gains from clinical LT in the home. To continue progress towards this goal,

the chair design should be refined to optimize for production of multiple units. In

addition, many of the suggested features discussed above are oriented towards making

the chair more useable at home, so they should be considered with any future redesigns.

Finally, any new designs will require further testing to ensure safety when used in the

home. A study to determine the minimum number of sensors that would still allow

tracking of chair use would also be appropriate.

To extend our knowledge of how use of rocking chairs in ABT contributes to the

progression of improved neuromuscular activation, studies involving long-term use of

rocking chairs in the home could be pursued. This kind of longitudinal research could

104

provide insight into whether rocking contributes to improvement in or retention of

intrinsic trunk control.

As mentioned previously, datasets representing more subjects, and representing

repeated data collections at different points in time would contribute to improved

training of machine learning models. Future research in this area could focus on creating

models that will generalize better, and on defining the reliability of models. In addition

to training with larger datasets, it would be helpful to work with therapists and other

clinicians to define the optimal balance between reliability of predictions, and specificity

of the predictions made.

Finally, other uses for the sensor data collected from the rocking chair could be

investigated. One example that is already in development is the incorporation of

advanced control mechanisms to use sensor input to assess the intent of the child. For

example, when a child needs additional help to maintain rocking, or alternatively needs

additional resistance so the rocking activity will continue to challenge them, mechanical

actuators working in concert with the sensor data could assist/resist rocking as

appropriate. Gamification of the rocking experience could also be explored using sensor

data to align the detection of activation of certain muscles and a reward system tied to

progress in the game.

6.5 Concluding Summary

In summary, this work has produced several valuable results. First, it has resulted

in a rocking chair that was designed and validated as a safe way for children with

impairments due to SCI to move independently, and to activate muscles. This has the

potential for use an integral component of a course of ABT therapy for children with SCI.

This study is also the first to investigate muscle activation patterns during

rocking, and to verify that rocking in a rocking chair can activate muscles throughout the

body for children with varying levels of impairment due to SCI. This sets the stage for

use of rocking chairs in the context of pediatric SCI, provides a base on which to build

further research into this topic and lays the groundwork for the development of new

techniques for quantifying muscle activation using common sensors combined with

powerful machine learning techniques. These tools should continue to be developed and

implemented in ways that will benefit children with SCI and, along with other therapies,

help them to improve their trunk control and their quality of life. Further research into

the benefits of rocking in this population should also continue to better define how these

tools can be used effectively, and to discover new applications where their benefit can be

realized.

Novel Rehabilitation Rocking Chair for Trunk Muscle Activation and Pattern Analysis in Children with Spinal Cord Injury.

Students also viewed