Argument

Artificial intelligence in healthcare: a review on predicting clinical needs Djihane Houfani, Sihem Slatnia, Okba Kazar , Hamza Saouli and Abdelhak Merizig

LINFI Laboratory, University of Biskra, Biskra, Algeria

ABSTRACT Artificial Intelligence is revolutionizing the world. In the last decades, it is applied in almost all fields especially in medical prediction. Researchers in artificial intelligence have exploited predictive approaches in the medical sector for its vital importance in the process of decision making. Medical prediction aims to estimate the probability of developing a disease, to predict survivability and the spread of a disease in an area. Prediction is at the core of modern evidence-based medicine, and healthcare is one of the largest and most rapidly growing segments of AI. Application of technologies such as genomics, biotechnology, wearable sensors, and AI allows to:

(1) increase availability of healthcare data and rapid progress of analytics techniques and make the foundation of precision medicine;

(2) progress in detecting pathologies and avoid subjecting patients to intrusive examinations; (3) make an adapted diagnosis and therapeutic strategy to the patient’s need, his

environment and his way of life.

In this research, an overview of applied methods on the management of diseases is presented. The most used methods are Artificial Intelligence methods such as machine learning and deep learning techniques which have improved diagnosis and prognosis efficiency.

ARTICLE HISTORY Received 6 March 2020 Accepted 26 December 2020

KEYWORDS Predictive medicine; artificial intelligence; prediction; healthcare; diagnosis; prognosis; breast cancer; cardiovascular diseases

1. Introduction

The healthcare domain is facing many challenges. In particular, handling large amounts of data (Big Data) will be a critical issue due to its sensibility. Also, these data are growing continuously and are some- times more complex which need diagnosis time and rising costs. In fact, every area has been impacting most healthcare providers and patients [1]. Predictive medicine is a field of medicine, which consists of determining the probability of disease. Its main role is to decrease the impact upon the patient such as by preventing mortality or limiting morbidity. Despite the several proposed solutions, medical prediction remains a challenging task and demands a lot of efforts. This is attributed to its vital importance in decision making. The main goals of predictive medi- cine are: (i) the practice of collecting and cataloguing characteristics of patients (big data analytics) [2]; (ii) analyzing that data to predict the patient’s individual risk for an outcome of interest; (iii) predicting which treatment in which individual will be most effective, and then intervening before the outcome occurs. Actually, Medical Informatics is at the junction of the disciplines of medicine and information technol- ogy and artificial intelligence tools. Both of these con- cepts play a crucial role in advancing the science of quality measurement. Artificial intelligence

technologies provide multiple services. They are used to improve accuracy, efficiency and public health, and maintain privacy and security of patient health information. The rest of the paper is organized as fol- lows. Section 2 introduces the predictive medicine domain. Section 3 describes some proposed works in medical prediction domain. Section 4 elaborates a comparative study of described works. Then, we finish with a discussion and a conclusion in Section 5.

2. Predictive medicine

Medicine is undergoing a revolution that virtualizes most medical practices. This revolution is emerging from the convergence of biology and medicine and computer technology with its ability to analyze ‘big data’ sets, deploy this information in business and social networks and create digital consumer devices measuring personal information [3].

Predictive medicine is a field of medicine that esti- mates the likelihood of disease occurring in the future taking into account relevant risk factors such as age, sex, clinical measured data, and so on. New technol- ogies allow to characterize infectious agents more rapidly and to produce effective vaccines more quickly. To generate predictive models of health and disease for each patient, researchers are developing

© 2021 Informa UK Limited, trading as Taylor & Francis Group

CONTACT Djihane Houfani [email protected] LINFI Laboratory, Department of Computer Science, University of Biskra, P.O. Box 145 RP, 07000 Biskra, Algeria

INTERNATIONAL JOURNAL OF HEALTHCARE MANAGEMENT 2022, VOL. 15, NO. 3, 267–275 https://doi.org/10.1080/20479700.2021.1886478

powerful new tools by exploiting artificial intelligence technology and biology techniques.

3. Literature review

Janghel et al. [4] developed a system for diagnosis, prognosis, and prediction of breast cancer (BC) using ANN models to assist doctors. Four models of neural networks were used to implement this system: Back Propagation Algorithm (MLP), Radial Basis Function Networks (RBF), Learning vector Quantiza- tion (LVQ), and Competitive Learning Network (CL). LVQ gave the best accuracy in the testing data set. However, the performed experiments of this work were limited to single database with a limited attri- butes for breast cancer.

Vikas and Saurabh [5] proposed diagnosis system for detecting BC based on three data mining tech- niques RepTree, RBF Network and Simple Logistic. These algorithms were used to predict the survivability rate of breast cancer data set. The three classification techniques were compared to find the most accurate one for predicting cancer survivability rate. The data used in this study were provided by the University Medical Centre, Institute of Oncology, Ljubljana, Yugoslavia. Authors used WEKA software to implement the machine learning algorithms.

The objective of Bichen et al. [6] in this research was to diagnose breast cancer by extracting tumor fea- tures. Authors developed a hybrid of K-means and SVM algorithms to extract useful information and diagnose the tumor. The K-means algorithm was uti- lized to recognize the hidden patterns of the benign and malignant tumors separately. Then, to obtain a new classifier an SVM was used.

Karabatak and Cevdet Ince [7] proposed an auto- matic diagnosis system based on associative rules (AR) and neuronal network for detecting breast can- cer. This method consisted of two stages. In the first stage, association rules were used to reduce the input feature vector dimension. Then, in the second stage neural network used these inputs and classified the breast cancer data. This method worked well; how- ever, it performs poorly if the features are not chosen well.

Seera and Lim [8] proposed a hybrid intelligent sys- tem based on Fuzzy Min–Max neural network, the Classification and Regression Tree, and the Random Forest (RF) model for undertaking medical data classification problems. This system had two impor- tant practical implications in the domain of medical decision Support: accuracy and the ability to provide explanation and justification for the prediction. The results were evaluated using three benchmark medical data sets.

Nilashi et al. [9] developed a knowledge-based sys- tem for the classification of breast cancer disease using

Expectation Maximization (EM), Classification and Regression Trees (CART), and Principal Component Analysis (PCA). The proposed system can be used as a clinical decision support system to assist medical practitioners in the healthcare practice.

Nguyen et al. [10] proposed a computer-aided diag- nostic system to distinguish benign breast tumor from malignant one. Their method consisted of two stages in which a backward elimination approach of feature selection and a learning algorithm RF are hybridized. The average obtained classification accuracy was between 99.70 and 99.82% in test phase applied for Wisconsin Breast Cancer Diagnosis Dataset (WBC- DD) and Wisconsin Breast Cancer Prognostic Dataset (WBCPD). This result indicated that the proposed method can be applied to other breast cancer pro- blems with different data sets especially with ones that have a higher number of training data. However, RF becomes slow and ineffective for real-time predic- tions when a large number of trees are generated.

Ahmed et al. [11] developed a Computer-Aided Diagnosis (CAD) scheme for the detection of breast cancer using deep belief network (DBN) unsupervised path followed by back propagation supervised path. The proposed system was tested on the Wisconsin Breast Cancer Dataset (WBCD) and gave an accuracy of 99.68%. However, this approach was computation- ally expensive.

Thein and Tun [12] proposed a breast cancer classification approach. This approach was based on the Wisconsin Diagnostic and Prognostic Breast Can- cer and the classification of different types of breast cancer datasets. The proposed system implemented the island-based training method to obtain better accuracy and less training time by using and analyzing between two different migration topologies. However, in this method same parameters may not guarantee the global optimum solution.

Arpit et al. [13] proposed a GONN algorithm, for solving classification problems. This algorithm was used to classify breast cancer tumors as benign or malignant. To demonstrate their results, authors took theWBCD database fromUCIMachine Learning repository and compared the classification accuracy, sensitivity, specificity, confusion matrix, ROC curves, and AUC under ROC curves of GONN with classical model and classical Back propagation model. How- ever, in this algorithm, only crossover and mutation operators were improved and it was applied only on WBCD database.

Dheeba et al. [14] proposed a new classification approach for the detection of breast abnormalities in digital mammograms using Particle Swarm Optimized Wavelet Neural Network (PSOWNN). The proposed work was based on extracting Laws Texture Energy Measures from the mammograms and classifying the suspicious regions by applying a pattern classifier

268 D. HOUFANI ET AL.

and applied to real clinical database. However, PSOWNN method suffers from difficulty in finding their optimal design parameters.

Raúl Ramos-Polĺan et al. [15] proposed and evalu- ated a method to design mammography-based machine learning classifiers (MLC) for breast cancer diagnosis. This method allowed to characterize breast lesions according to BI-RADS classes (grouped by benign and malignant). This approach gave a good accuracy but it was evaluated on one database.

Geert Litjens et al. [16] explored deep learning to improve the objectivity and efficiency of histopatholo- gic slide analysis. Authors used convolutional neural network to digitized histopathology through two different experiments: prostate cancer detection in hematoxylin and eosin (H&E)-stained biopsy speci- mens and identification of metastases in sentinel lymph nodes obtained from breast cancer patients. This method gave accurate results but it showed some detection errors in the prostate cancer exper- iment and data were extracted from a single center. This approach was performing in terms of accuracy but it was computationally expensive.

Wang et al. [17] proposed a deep learning based approach for detecting metastatic breast cancer from whole slide images of sentinel lymph nodes. This approach was tested on Camelyon16 dataset. The pro- posed approach improved in the reproducibility, accu- racy, and clinical value of pathological diagnoses; however, it was computationally expensive.

Gonźalez-Briones et al. [18] designed a multi- agents based system to manage information of expression arrays. In this system, different data mining techniques and databases were used to analyze expression profiles; its aim was to provide genes that show differences between samples from younger and older patients to discover why older women respond better to the treatment. The system identified the genes that can be therapeutic targets. However, for a best result, it is necessary to check if the gene in ques- tion is over or under-expressed.

Cruz-Roa et al. [19] proposed a deep learning based tool that employed a convolutional neural network (CNN) to detect automatically presence of invasive tumors on digitized images. This approach was tested on data from different sources. However, while using this method, some breast cancer regions were incor- rectly classified.

In this paper, Ankur and Jaymin [20] proposed a predictive model for heart disease detection using Machine Learning and Data Mining techniques. The proposed approach combined between Naive Bayes (NB) and Genetic Algorithm (GA) to classify heart diseases. Data were collected from Cleveland Heart Disease Data set (CHDD) available on the UCI Repo- sitory. Nonetheless, this model could not predict specific heart disease.

In this paper, Vignon-Clementel et al. [21] pro- posed a 3D simulation approach for blood flow and arterial pressure, this method has been applied to cal- culate hemodynamic quantities in various physiologi- cally relevant cardiovascular models, including patient-specific examples, to study non-periodic flow phenomena, often seen in normal subjects and in patients with acquired or congenital cardiovascular disease. However, it was difficult to measure pressures and flow rates in vivo simultaneously and it was feas- ible in a very limited number of research cases. Fur- thermore, the vessel wall displacements were overestimated because of resistance boundary condition.

In this paper, Subanya et al. [22] used meta-heuris- tic algorithm (bee colony) to determine the subset of optimal characteristics with better classification accu- racy in the diagnosis of cardiovascular disease. Data were taken from UCI repository (a database of cardi- ovascular diseases).

Shaikh et al. [23] used ANNs to predict the medical prescription of heart disease. This work included detailed information about the patient’s symptoms and the pretreatment that was done. Doctors can also use this web-based tool for the diagnosis of heart disease using the basic radial function. Outputs of this system have been compared with the prescrip- tions of the doctors and it was satisfactory.

In this paper, Singh et al. [24] applied Structural Equation Model (SEM) to identify the strength of relationships among variables that are considered related to the cause of Cardiovascular Diseases (CVDs) and Fuzzy Cognitive Map (FCM) to evaluate obtained results in a predictive system that helps for the detection of people who are at risk of developing CVDs. In this study, data have been extracted from Canadian Community Health Survey (CCHS) data source. However, authors did not use enough attri- butes to have a very accurate model.

Singh et al. [24] proposed a predictive system of CVDs using quantum neural network (QNN) for machine learning. Data were extracted from 689 patients showing symptoms of CVD and the dataset of 5209 CVD patients of the Framingham study. This system had been experimentally evaluated and compared with Framingham risk score (FRS). This proposed system predicted the CVD risk with high accuracy and was able to update itself with time.

In this paper, Venkatalakshmi et al. [25] designed and developed diagnosis and prediction system for heart diseases. In this system, prediction was based on two algorithms: DT and NB were executed on Weka tool; dataset consisted of attributes and values which are collected from UCI machine learning repo- sitory which is a repository of databases, domain the- ories, and data generators. In order to improve the efficiency and accuracy, an optimization process

INTERNATIONAL JOURNAL OF HEALTHCARE MANAGEMENT 269

genetic algorithm has been used. In this system, a large amount of data were used that must be reduced and take into consideration only subset of attribute sufficient for heart disease prediction.

Boden et al. [26] proposed a mathematical method to predict the probability of surgery prior to the first visit based on a sample of 8006 patients with low back pain. Independent risk factors for undergoing spinal surgery were identified by using univariate and multivariate statistical analysis, and the Spine Sur- gery Likelihood (SSL) model was created using a ran- dom sample of 80% of the total patients in the used cohort, and validated on the remaining 20%. However, this method was unable to track patients who have undergone surgery in a different facility and, therefore, may have been misclassified in the non-surgical group.

In this paper, Søreide et al. [27] proposed an approach that used Artificial Neural Network (ANN), multilayer perceptron (MLP) to predict the mortality of patients with perforated peptic ulcer. Input to this approach was a sample of patients ana- lyzed by Statistical Package for Social Sciences (IBM SPSS v. 21, Inc. for Mac). Its principle was to propose three models of MLP and give the model with the opti- mal performance. However, in this kind of approaches, the intervention of the human expert is essential for the collection of data and garbage-in, gar- bage out problem can exist.

Nyssa et al. [28] proposed, in their article, a predic- tive model of rabies in Tennessee; it was based on spatial analysis. The proposed method consisted of:

(1) Data acquisition from the Tennessee’s Health Department

(2) Data processing using ArcGIS software to get the predictive model

(3) Spatial analysis using Fragstats and Circuitscape software.

Result of this system was a set of models (maps) such as distribution models, density model and so

on. However, it did not allow a real-time disease’s sur- veillance and was not efficient in case of companies with large population.

In this paper, Sharmila Devi et al. [29] described in this paper a distributed system of e-health for the automatic diagnosis of the situation of a patient based on his data without the participation of a doctor. This service was provided on the Internet. When a patient’s situation changes, the system will automati- cally alert the doctor. This has been implemented using Multi-Agent System (MAS) and Adaptive Neuro-Fuzzy Inference System (ANFIS). The different agents in the system were in different places and used an asynchronous communication to communicate each other.

In this paper, Kaberi et al. [30] presented an approach that consisted of hybridization between GA, harmony search algorithms (HAS) and support vector machine (SVM) for the selection of informative genes. However, heuristic methods depend on the pro- blem and they are generally based on a local optimum that fails to obtain the optimal overall solution.

Golnaz et al. [31], in this paper, proposed a feature selection method based on a genetic algorithm. To evaluate the subsets of the selected characteristics, the k nearest neighbors (KNN) classifier was used and validated on a set of data of the UCI database.

In this paper, Talayeh et al. [32] used unbalanced classification techniques: NB, Radial Basis Function Neural Network (RBFNN), 5-Nearest Neighbors, Decision Trees (DT), SVMs, and Logistic Regression (LR) to identify the complications of bariatric surgery for each patient. The combination of classification methods made possible to achieve higher performance measures (Figure 1).

3.1. Breast cancer prediction and diagnosis

In this section, we discuss researches which used different AI methods to manage breast cancer disease. Table 1 summarizes the reviewed work dealing with

Figure 1. Flow diagram that summarizes the reviewed researches.

270 D. HOUFANI ET AL.

Table 1. Summary table of researches which used different techniques to manage breast cancer disease. Works Objective Method Data Result Limitations

Janghel et al. [4] Diagnosis (malignant and benign cells classification)

ANN (application of 4 methods) WBCD (collected data) Best classification method (LVQ)

Use of one dataset with limited attributes

Chaurasia et al. [5]

Diagnosis/prognosis (survivability prediction)

Data mining (Rep tree, RBF network, simple logistic)

University Medical Centre, Institute of oncology Ljubljana Yugoslavia

Best method (simple logistic) Use of one dataset with limited attributes

Zheng et al. [6] Diagnosis K-means and SVM classifier WBCD (table: attributes-values) Features selection for tumors classification

It is not implemented in a large-scale sparse data set

Karabatak et al. [7]

Diagnosis AR and neural network WBCD (table: attributes-values) Tumors classification Applied on one dataset

Seera et al. [8] Medical data classification FMin-MaxNN, Classification and Regression Tree, RF model

WBCD, Pima Indians Diabetes, and Liver Disorders from the UCI Repository of Machine Learning

Undertaking medical data classification problems

Good

Nilashi et al. [9] Diagnosis - EM for data clustering - Fuzzy logic for data classification - PCA to solve multi-collinearity problem - CART for automatic fuzzy rules generation

- WBCD (table: attributes-values) - Mammographic mass dataset

Tumors classification EM fails on high-dimensional data sets due to numerical precision problems

Nguyen et al. [10] Diagnosis and prognosis Feature selection RF classifier

WBCDD and WBCPD Tumor classification RF becomes slow and ineffective for real-time predictions when a large number of trees are generated

Abdel-Zaher et al. [11]

Diagnosis DBN (unsupervised) for pre- training Supervised back propagation for classification

WBCDD Tumor classification Computationally expensive

Thein et al. [12] Diagnosis Differential evolution algorithm (for training) Parallelism

WBCDD A neural network for Tumor Classification

Same parameters may not guarantee the global optimum solution

Bhardwaj et al. [13]

Diagnosis GONN WBCDD Tumor classification - Only crossover and mutation operators are improved - Applied on one dataset

Dheeba [14] Diagnosis PSOWNN Mammogram screening center (real data `a images) BC detection - Dependency on initial point and parameters. - Difficulty in finding their optimal design parameters

Ramos-Polĺan et al. 15]

Diagnosis Machine learning classifier BCDR ML classifiers Evaluated on one database

Litjens et al. [16] Diagnosis Deep learning (CNN) Collected patient’s specimens Histopathologic slide analysis Computationally expensive Wang et al. [17] Diagnosis Deep learning Camelyon16 dataset Cancer metastases

identification Computationally expensive

Gonźalez-Briones et al. [18]

Prognosis MAS Deep learning

Samples provided by Salamanca Cancer Institute Gene selection Computationally expensive

Cruz-Roa et al. [19]

Diagnosis CNN Digital images from different institutions Invasive breast cancer classification

Some errors of classification

IN TERN

A TIO

N A L JO

U RN

A L O F H EA

LTH C A RE

M A N A G EM

EN T

271

breast cancer disease. The first column refers to the investigated work; the second column is the objective of the work; the third column is the used method to handle the disease; the fourth column refers to the used dataset of the paper; the fifth column consists of the results; and finally, the last one refers to the limitations of the proposed work.

3.1.1. Discussion Breast cancer is the most common cause of women’s deaths worldwide [33]. It is a result of mutations, anarchic division, and abnormal changes of cells.

AI applies algorithms on a large volume of health- care data to assist clinical practice. These algorithms show their ability to improve accuracy by learning and self-correcting.

After observing the reviewed researches that man- age breast cancer disease, we can notice that machine learning techniques are widely used in diagnosis, tumors classification and breast cancer prediction to assist physicians in decision making process and early detection. The most used dataset is WBCD from UCI Repository. These works show a good per- formance in terms of accuracy. However, some techni- cal problems can be considered:

(1) Computational and memory expenses (2) Data availability: Training AI systems requires

large amounts of structured and comprehensive data. However, the available data are fragmented,

incomplete and unstructured, these problems increase the risk of error

(3) Overfitting problem: This occurs when the model properly fits the training data and encounters difficulties for generalization on new or unseen data (validation data).

(4) Reproducibility issue: A study is reproducible when others can replicate the results using the same algorithms, data and methodology.

3.2. Other diseases

Researches mainly concentrate around diseases which are leading causes of death. We can classify them into the following types: cardiovascular disease, cancers, viral disease, and nervous system disease; therefore, early diagnosis and prognosis are fundamental to pre- vent the deterioration of patients’ health status.

Table 2 summarizes the reviewed work dealing with different diseases. The first column refers to the inves- tigated work; the second column is the tackled disease; the third column is the used method to handle the dis- ease; the fourth column refers to the objective of the paper; the fifth column consists of the used dataset; and finally, the last one refers to the achieved perform- ance of the proposed work.

3.2.1. Discussion The use of artificial intelligence techniques in medical prediction to manage different diseases shows a

Table 2. Summary table of researches which used different techniques to manage multiple diseases. Works Disease Method Objectives Input Performance

Makwana et al. [20] CVD ML and Data Mining Heart disease detection Cleveland Heart Disease Data set

Good but it can be improved

Vignon et al. [21] Cardiovascular system

- Mathematic equation - Analog electrical circuit

3D simulation approach for blood flow and arterial pressure

Measured data Its validation is proven in vitro and in vivo data

Subanya et al. [22] CVD Meta-heuristic algorithm (bee colony)

CVD Classification UCI repository Good

Hannan et al. 23] CVD ANN Medical prescription of heart disease prediction

Patient information Good

Singh et al. [24] Cardiovascular disease

SEM and FCM Building a Cardiovascular Disease Predictive Model

CCHS dataset It can be improved

Narain et al. [25] CVD QNN Risk of CVDs prediction Patients with CVDs Good but it can be improved

Venkatalakshmi et al. [26]

CVD DT and NB Heart diseases prediction Attributes and values from UCI database

Good but it can be improved

Boden et al. [27] Orthopedic surgery Mathematical method Surgery prior’s probability prediction

Patient-reported data Low level of evidence (4)

Søreide et al. [28] Gastric disease ANN modeling Mortality prediction for patients with

Gastric disease ANN modeling

Nyassa et al. [29] Viral disease spatial analysis Rabies prediction in Tennessee

Tennessee’s Health Department

Good accuracy

Devi et al. [30] Neck and arm pain disease

MAS and ANFIS Patients automatic diagnosis Patient-reported data Good

Das et al. [31] Informative genes selection

GA, HAS and SVM Selection of informative genes Gene expression dataset

Good

Sahebi et al. [32] Feature selection method

GA Feature selection and classification optimization

UCI Arrhythmia database,

Good

Razzaghi et al. [33] Bariatric surgery Imbalanced classification techniques

Identify bariatric surgery’s complications

The Premier Healthcare Database

Good

272 D. HOUFANI ET AL.

performance improvement in terms of accuracy, speed and interoperability. Machine Learning techniques are suitable for the management of multiple diseases (Figure 2). Furthermore, their use makes disease man- agement more reliable by reducing diagnosis and therapeutic errors, and extracting useful information from large amount of data to predict health outcomes. Multiple data are used in these researches such as medical images, patient’s reported, data datasets from UCI Repository and several public datasets.

3.3. Application of AI in healthcare: General challenges

This paper shows that Artificial intelligence brings important developments to health-care field, however, a subsequent research challenges remaining:

(1) Data quality and availability: Acquiring large amounts of high-quality clinical datasets is a very difficult process, because they are in multiple formats and fragmented across different systems and generally have limited access [34].

(2) Security and privacy issue: Several researchers have been interested in this concept and have pro- posed work to manage data security [35] because it is one of the biggest challenges facing AI sys- tem’s developers. The requirement of large amounts of data from many patients may affect their data privacy.

(3) Bias issue: AI systems learn to make decisions based on training data which can include biases.

(4) Computational cost: Most reviewed works are computationally expensive, which is not beneficial for both clinician and patient.

(5) Interpretability: The most important task in the healthcare domain is evaluating and validating the proposed approach to be accepted by the community.

(6) Injuries and error: An AI system may be some- times wrong by failing in diseases prediction or in a drug recommendation or in predicting the response of a patient to a specific treatment.

These failures can occur patient injury or other healthcare problems.

4. Conclusion

Medical prediction is a very important challenge for clinicians because it has a direct influence on their daily practice. In the last decade, the death rate increases significantly, this required methods and tools for accurate and early detection of diseases. While going through literature review, we noticed that researchers are interested in medical prediction especially in the diagnosis and prognosis of breast can- cer using methods and approaches of artificial intelli- gence such as ANN, deep learning and data mining, and so on. The authors in the literature proposed sys- tems and compared them to other existing works. We can note that their approaches are efficient in terms of accuracy; however, most of them are time-consuming in the training phase. We can also notice that very few of these research works have actually been integrated the clinical practice.

In this paper, we discussed the biggest challenges facing the application of AI in the healthcare field. To handle these challenges, several solutions can be proposed:

(1) High-quality data generation and availability: To build an efficient AI system it is important to pro- ceed on good datasets, that’s why it is important to create high-quality databases accessible by researchers and AI systems developers in a man- ner consistent with protecting patient privacy. Blockchain technology can be used to secure per- sonal and medical data [36].

(2) Quality supervising: Good training and validating of AI systems will help address the risk of errors and patient injury.

(3) Good exploitation of AI methods: Hybridization of deep learning method with optimization algor- ithms [37], parallelization, could be powerful for time and cost reduction. Big data analytics also offers several opportunities in this field [38].

The used techniques in reviewed works include mathematical methods, evolutionary computing, case-based reasoning, fuzzy logic, ANNs, data mining, machine learning, deep learning, and intelligent agents. However, the medical prediction is not wide- spread due to several constraints. Hence comprehen- sive research needs to be done in this sphere keeping an eye towards developing hybrid techniques that could be employed to predictive medicine. The selec- tion of the appropriate technique is important for developing and implementing disease diagnosis sys- tems. As a perspective of this work, we aim to designFigure 2. Used techniques in medical literature.

INTERNATIONAL JOURNAL OF HEALTHCARE MANAGEMENT 273

our medical predictive approach based on deep reinforcement learning and genetic algorithms to improve breast cancer diagnostic performance. Fur- thermore, to overcome big data problems, the number of characteristics in the dataset must be reduced which allows ensuring the quality of data (QoD). The advan- tage of developing deep learning technique for the management of breast cancer disease will be reached by applying it as support tools that help physicians in diagnosis, prognosis, and treatment. By using this type of systems reading variability by physicians will be eliminated. Besides, more quick and accurate diag- nosis will result.

Despite the several challenges facing AI application in healthcare field, it is very promising in decision- making aid, physician and patient medical support, and prediction and we believe there are still significant perspectives on this topic.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes on contributors

Djihane Houfani received the Licence andMaster degrees in Computer Science from University of Biskra, Algeria in 2015 and 2017, respectively. She is now a PhD student in artificial intelligence at the University of Biskra and her cur- rent research interest includes medical prediction, deep learning, multi-agent systems and optimization.

Sihem Slatnia was born in the city of Biskra, Algeria. She followed her high studies at the university of Biskra, Algeria at the Computer Science Department and obtained the engineering diploma in 2004 on the work “Diagnostic based model by Black and White analyzing in Background Petri Nets”, After that, she obtained Master diploma in 2007 (option: Artificial intelligence and advanced system’s information), on the work “Evolutionary Cellular Automata Based-Approach for Edge Detection”. She obtained PhD degree from the same university in 2011, on the work “Evol- utionary Algorithms for Image Segmentation based on Cel- lular Automata”. Presently she is an associate professor at computer science department of Biskra University. She is interested to the artificial intelligence, emergent complex systems and optimization.

Okba Kazar professor in the Computer Science Department of Biskra, he helped to create the laboratory LINFI at the University of Biskra. He is a member of international con- ference program committees and the “editorial board” for various magazines. His research interests are artificial intel- ligence, multi-agent systems, web applications and infor- mation systems.

Hamza Saouli received the Master and Doctorate degrees in Computer Science from University of Mohamed Khider Biskra (UMKB), the Republic of Algeria in 2010 and 2015, respectively. He is a university lecturer since 2015 and his research interest includes artificial intelligence, web services and Cloud Computing.

Abdelhak Merizig obtained his Master degree by 2013 from Mohamed Khider University, Biskra, Algeria, He is working

on an artificial intelligence field. He obtained his PhD degree from the same university in 2018. Abdelhak Merizig is now a university lecturer at the computer science depart- ment of Biskra University. Also, he is a member of LINFI Laboratory at the same University. His research interest includes multi-agent systems, service composition, Cloud Computing and Internet of Things.

ORCID

Okba Kazar http://orcid.org/0000-0003-0522-4954

References

[1] Usman Ahmad M, Zhang A, Goswami M, et al. A pre- dictive model for decreasing clinical no-show rates in a primary care setting. Int J Healthcare Manag. 2019;11:1–8.

[2] Kamble SS, Gunasekaran A, Goswami M, et al. A sys- tematic perspective on the applications of big data analytics in healthcare management. Int J Healthcare Manag. 2018;12:226–240.

[3] Hood L, Flores M. A personal view on systems medi- cine and the emergence of proactive P4 medicine: pre- dictive, preventive, personalized and participatory. New Biotechnol. 2012;6(23):613–624.

[4] Janghel RR, Shukla A, Tiwari R, et al. Breast cancer diagnosis using artificial neural network modelsThe 3rd International Conference on Information Sciences and Interaction Sciences 2010.

[5] Chaurasia V, Pal S. Data mining techniques: to predict and resolve breast cancer survivability. Int J Comput Sci Mob Com. 2014;3(1):10–22.

[6] Zheng B, Yoon SW, Lam SS. Breast cancer diagnosis based on feature extraction using a hybrid of K- means and support vector machine algorithms. Expert Syst Appl. 2013;41:1476–1482.

[7] Karabatak M, Ince MC. An expert system for detection of breast cancer based on association rules and neural network. Elsevier, Expert Syst Appl. 2009;36:3465–3469.

[8] Seera M, Lim CP. A hybrid intelligent system for medical data classification. Expert Syst Appl. 2013;41:2239–2249.

[9] Nilashi M, Ibrahim O, Ahmadi H, et al. A knowledge- based system for breast cancer classification using Fuzzy logic method. Telemat Inform. 2017;34:133– 144.

[10] Nguyen C, Wang Y, Nguyen HN. Random forest clas- sifier combined with feature selection for breast cancer diagnosis and prognostic. J Biomed Sci Eng. 2013;06:551–560.

[11] Abdel-Zaher AM, Eldeib AM. Breast cancer classifi- cation using deep belief networks. Expert Syst Appl. 2015;46:139–144.

[12] Thein HTT, Tun KMM. An approach for breast can- cer diagnosis classification using neural network. Adv Com Int J. 2015;6:1–11.

[13] Bhardwaj A, Tiwari A. Breast cancer diagnosis using genetically optimized neural network model. Expert Syst Appl. 2015;42:4611–4620.

[14] Dheeba J, Singh NA, Selvi ST. Computer-aided detec- tion of breast cancer on mammograms: a swarm intel- ligence optimized wavelet neural network approach. J Biomed Inform. 2014;49:45–52.

[15] Ramos-Polĺan R, Guevara-Ĺopez MA, Súarez-Ortega C, et al. Discovering mammography-based machine

274 D. HOUFANI ET AL.

learning classifiers for breast cancer diagnosis. J Med Sys. 2012;36:2259–2269.

[16] Litjens G, Śanchez CI, Timofeeva N, et al. Deep learn- ing as a tool for increased accuracy and efficiency of histopathological diagnosis. Sci Rep. 2016;6:1–11.

[17] Wang D, Khosla A, Gargeya R, et al. Deep learning for identifying metastatic breast cancer. Int Symp Biomed Imaging. 2016: 1–6.

[18] Gonźalez-Briones A, Ramos J, De Paz JF, et al. Multi- agent system for obtaining relevant genes in expression analysis between young and older women with triple negative breast cancer. J Integr Bioinform. 2015;12:1–14.

[19] Cruz-Roa A, Gilmore H, Basavanhally A, et al. Accurate and reproducible invasive breast cancer detection in whole slide images: a deep learning approach for quantifying tumor extent. Sci Rep. 2017;7:1–14.

[20] Makwana A, Patel J. Decision support system for heart disease prediction using data mining techniques. Int J Comput Appl. 2015;117 (22):1–5.

[21] Vignon-Clementel IE, Figueroa CA, Jansen KE, et al. Outflow boundary conditions for 3D simulations of non-periodic blood flow and pressure fields in deformable arteries. Comput Methods Biomech Biomed Engin. 2010;13:625–640.

[22] Subanya B, Rajalaxmi RR. Feature selection using artificial Bee colony for cardiovascular disease classification 2014 International Conference on Electronics and Communication Systems (ICECS) IEEE. 2014;1–6.

[23] Hannan SA, Mane AV, Manza RR, et al. Prediction of heart disease medical prescription using radial basis function. Comput Intell Comput Res. 2010;15:1–6.

[24] Singh M, Martins LM, Joanis P, et al. Building a cardi- ovascular disease predictive model using structural equation model Fuzzy cognitive map 2016 IEEE International Conference on Fuzzy Systems (FUZZ- IEEE). 2016.

[25] Narain R, Saxena S, Goyal AK. Cardiovascular risk prediction: a comparative study of Framingham and quantum neural network based approach. Patient Prefer Adherence. 2016;10:1259–1270.

[26] Venkatalakshmi B, Shivsankar MV. Heart disease diagnosis using predictive data mining. Int J Innov Res Sci Eng Technol. 2014;3(3):1873–1877.

[27] Boden LM, Boden SA, Premkumar A, et al. Predicting likelihood of surgery prior to first visit in patients with back and lower extremity symptoms: a simple math- ematical model based on over 8000 patients. SPINE J. 2018;43:1–27.

[28] Søreide K, Thorsen K, Søreide JA. Predicting out- comes in patients with perforated gastroduodenal ulcers: artificial neural network modelling indicates a highly complex disease. Eur J Trauma Emergency Surgery. 2015;41:91–98.

[29] Nyssa H, Wilson TP, Andrew C. Spatiotemporal analysis and predictive modeling of rabies in Tennessee. J Geogr Inf Sys. 2018;10:89–110.

[30] Devi CS, Ramani GG, Pandian JA. Intelligent E- healthcare management system in medicinal science. Int J PharmTech Res. 2014;6:1838–1845.

[31] Das K, Mishra D, Shaw K. A metaheuristic optimiz- ation framework for informative gene selection. Inform Med Unlocked. 2016;4:10–20.

[32] Sahebi G, Majd A, Ebrahimi M, et al. A reliable weighted feature selection for auto medical diagnosis. IEEE 15th International Conference on Industrial Informatics (INDIN). 2017; 985–991.

[33] Razzaghi T, Safro I, Ewing J, et al. Predictive models for bariatric surgery risks with imbalanced medical datasets. School Computing. 2017;280:1–18.

[34] U.S. Cancer Statistics Working Group. United States Cancer Statistics: 19992008 Incidence and Mortality Web-based Report. Atlanta (GA): Department of Health and Human Services, Centers for Disease Control and Prevention, and National Cancer Institute; 2012.

[35] Kamble SS, Gunasekaran A, Goswami M, et al. A sys- tematic perspective on the applications of big data analytics in healthcare management. Int J Healthcare. 2018: 1–15.

[36] Sarkara BK, Sana SS. A conceptual distributed frame- work for improved and secured healthcare system. Int J Healthcare. 2018: 1–14.

[37] Attaran M. Blockchain technology in healthcare: chal- lenges and opportunities. Int J Healthc Manag. 2020;13:1–14.

[38] Bahadori M, Hosseini SM, Teymourzadeh E, et al. Mehdi Raadabadi Khalil Alimohammadzadeh. a sup- plier selection model for hospitals using a combi- nation of artificial neural network and fuzzy VIKOR. Int J Healthc Manag. 2017;13:1–9.

INTERNATIONAL JOURNAL OF HEALTHCARE MANAGEMENT 275

Copyright of International Journal of Healthcare Management is the property of Taylor & Francis Ltd and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.