Computational intelligence approaches for classification of medical data: State-of-the-art, future challenges and research directions

Neurocomputing 276 (2018) 2–22

Contents lists available at ScienceDirect

Neurocomputing journal homepage: www.elsevier.com/locate/neucom

Computational intelligence approaches for classification of medical data: State-of-the-art, future challenges and research directions Ali Kalantari a, Amirrudin Kamsin a, Shahaboddin Shamshirband b,c,∗, Abdullah Gani a, Hamid Alinejad-Rokny d, Anthony T. Chronopoulos e,f a

Faculty of Computer Science & Information Technology, University of Malaya, 50603 Kuala Lumpur, Malaysia Department for Management of Science and Technology Development, Ton Duc Thang University, Ho Chi Minh City, Vietnam Faculty of Information Technology, Ton Duc Thang University, Ho Chi Minh City, Vietnam d UNSW Australia, Australia e Department of Computer Science, University of Texas, San Antonio, TX 78249, USA f Department of Computer Science (Visiting Faculty), University of Patras, Greece b c

a r t i c l e

i n f o

Article history: Received 23 August 2016 Revised 29 December 2016 Accepted 4 January 2017 Available online 19 September 2017 Keywords: Computational intelligence Medical application Big data Detection Ensemble algorithm

a b s t r a c t The explosive growth of data in volume, velocity and diversity that are produced by medical applications has contributed to abundance of big data. Current solutions for efficient data storage and management cannot fulfill the needs of heterogeneous data. Therefore, by applying computational intelligence (CI) approaches in medical data helps get better management, faster performance and higher level of accuracy in detection. This paper aims to investigate the state-of-the-art of computational intelligence approaches in medical data and to categorize the existing CI techniques, used in medical fields, as single and hybrid. In addition, the techniques and methodologies, their limitations and performances are presented in this study. The limitations are addressed as challenges to obtain a set of requirements for Computational Intelligence Medical Data (CIMD) in establishing an efficient CIMD architectural design. The results show that on the one hand Support Vector Machine (SVM) and Artificial Immune Recognition System (AIRS) as a single based computational intelligence approach were the best methods in medical applications. On the other hand, the hybridization of SVM with other methods such as SVM-Genetic Algorithm (SVM-GA), SVM-Artificial Immune System (SVM-AIS), SVM-AIRS and fuzzy support vector machine (FSVM) had great performances achieving better results in terms of accuracy, sensitivity and specificity. © 2017 Elsevier B.V. All rights reserved.

1. Introduction The era of big data has begun, due to large-volume, complex and growing number of data sets which are produced by various sources such as Internet of Things (IoT), government records, health records, multimedia, phone logs, social media and other digital traces [1–3]. Moreover, big data are being used to transform medical practice, inform business decision making, and modernize public policy [4]. Accordingly, the produced number of complex data from medical and healthcare increase rapidly with numerous essential information. Therefore, big data has infinite potential in efficiently storing, processing, querying, and analyzing medical data [5]. For instance, the Ayasdi organization provides information to the Mount Sinai Medical Center in the U.S. about the genetic sequencing of Escherichia coli (E. coli) bacteria. This data is ∗ Corresponding author at: Department for Management of Science and Technology Development, Ton Duc Thang University, Ho Chi Minh City, Vietnam. E-mail address: [email protected] (S. Shamshirband).

https://doi.org/10.1016/j.neucom.2017.01.126 0925-2312/© 2017 Elsevier B.V. All rights reserved.

used to investigate the resistance of antibiotics to bacterial strains. It utilizes topological data analysis which is a contemporary research methodology that can comprehend data characteristics [6]. In addition, there are various areas in healthcare industry including medical imaging [7–9], patient genomics [10,11], electronic health records (EHRs) [12,13], unstructured text data [14,15] and device, log and sensor data [16,17] that can have potential benefits from big data techniques and infrastructure [18]. However, there have been issues like security, privacy, the effectiveness of analysis and data quality which are very important in medical data and applications. Therefore, medical information shows valuable intellectual property and their usage is highly guarded which means the information management should not only be constrained on practices and established laws, but also by subjects expectation of privacy [19]. For example, a framework was proposed by Bertino et al. [20] to achieve the privacy and copyright protection for outsourced medical data by combining the techniques of binning and digital watermarking. Moreover, global disease network examination, using biological databases and pa-

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

tient data from electronic medical records (EMR), has emerged as a powerful modality for understanding the complexity of disease relationships [21,22]. In this study [23] the authors proposed the combination of different powerful approaches to compare disease network structure and connectivity between three different populations such as Hispanic/Latino, Caucasian and African American. Additionally, there are other challenges of big data in healthcare that are addressed by the procurement of information from complex heterogonous patient sources. These challenges include obtaining clinical notes and comprehending them in the right context, organizing medical imaging data, perceiving data concerning biomarkers and understanding large amounts genomic data which can be useful in clinical settings when the patient is assessed. Also, data about the patient’s psychological, behavioral, and social patterns can be accessed by various sensors [24]. A recent survey by Fang et al. [25] reviewed the current challenges, techniques and future directions for computational health informatics in big data. They summarized the challenges into four dimensions (Vs) of big data as volume, velocity, veracity and variety. In another recent investigation [5] the authors considered 6Vs (volume, velocity, veracity, variety, variability and value) possible challenges with complexity in big data. However, the recent technologies like Artificial Intelligence (AI) can assist in solving different complex issues. AI in its widest sense would demonstrate the capability of a machine to perform tasks similar to the human thought. Thus, AI has been used for computer systems with ability of task execution which is more complex than simple programing [26]. Artificial intelligence is also increasing in three V’s (volumes, velocities and variety of data) similar to big data. AI enables learning, delegation of difficult pattern recognition and additional responsibilities to computer-based methods under circumstances of large volumes of data. Furthermore, it helps velocity of data, by assisting fast computer-based decisions that make possible different choices. Besides, the variety issue is not solved only by parallelizing and distributing the problems. Variety is mitigated by capturing, structuring, and understanding unstructured data using AI and different analytics [27]. Therefore, AI and big data analytics could reshape the healthcare system with greater performance, provide healthcare insights, and improve the overall processes in two core elements for improving the productivity, efficiency and the quality of care [28]. With the intention of mitigating the performance of AI, the computational intelligence (CI) techniques adapt to medical data. CI typically refers to the ability of computer to learn a particular task from experimental observation or data, which facilitate the intelligent behavior in complex problems and changing environments [29]. CI techniques are classified based on single and hybrid methods, where single methods refer to those studies which use only one of the machine learning techniques (i.e. genetic algorithm (GA), particle swarm optimization (PSO), artificial immune system (AIS) and artificial neural network (ANN)) as a main method and the other classification refers to those studies that used hybridization of each two (or more than two) methods like Neuro-Fuzzy NF) and Fuzzy Support Vector Machine (FSVM). For instance, Latifogˇ lu et al. [30] used only artificial immune recognition system (AIRS) as the main technique for atherosclerosis diagnosis from carotid artery Doppler signals, which is considered as a single method classification. In another study by Gu et al. [31], for hybrid classification they used FSVM technique in medical datasets classification. The aim of this study is to investigate the state-of-the-art of computational intelligence (CI) approaches in medical data and to classify the CI techniques (single and hybrid) in terms of chronological design. Also, the study’s aim is to analyze the CI methods in terms of accuracy, sensitivity and specificity for medical and health informatics fields.

3

The rest of the paper is organized as follows. Section 2 provides the survey methodology. Section 3 presents the datasets used for CI approached applied to medical data. Section 4 defines the criteria for evaluation. Section 5 presents the state-of-the-art of single and hybrid CI approaches in medical data and presents the comparison based on the evaluation criteria. Lastly, Section 6 draws the overall conclusions.

2. Methodology In this study, 71 articles related to CI techniques in medical data were reviewed and selected from highly cited publications and credible sources as: Science Direct, IEEE, Springer and Web of Science (WoS). This paper integrates different classification of CI techniques which are used for medical data. Table 1 provides the list of literature works dealing with single and hybrid CI methods. The list of articles is given as a general overview of single and hybrid methods in terms of characteristics and current challenges of CI in medical data and healthcare. The table contains 2 horizontal sections (single and hybrid methods) and 4 vertical divisions which are single (or hybrid), classifier type, title of paper, and aim.

3. Datasets used for computational intelligence approaches in medical data To make the evaluation of performance for each approach for detection, diagnosis, prediction, analyzing and classification, it is required that the related datasets define the level of accuracy, sensitivity and specificity. Table 2 indicates the classification of the datasets which were for the most part obtained from University of California at Irvine (UCI) Machine Learning Repository, which is considered as a public dataset. Different types of medical datasets were used by researchers for coronary heart disease, ovarian, hepatitis, lung and breast cancer. For instance, the hepatitis disease dataset (including 19 attributes, and 155 samples based on two categories: 32 for die cases and 123 for live cases) was used to predict the disease of hepatitis [68]. In addition, medical images are one type of digital information data which are increasing. In these study [108,110] the authors used the public medical database of ImageCLEF 2007 in order to achieve a high classification rate.

4. Criteria used for evaluation The effectiveness of CI techniques in medical data is evaluated on how capable the single and hybrid methods are in making correct diagnosis, detection, monitoring and prediction in terms of accuracy (i.e. the overall proportion of correct classification) and sensitivity (i.e. the proportion of the positives correctly recognized) and specificity (i.e. proportion of the negatives correctly recognized). Table 3 indicates the analysis of three key aspects (accuracy, sensitivity and specificity) in medical data using computational intelligence approaches. The significance of the performance is emphasized, this is because medical data represents valuable intellectual property. As an example, in [67] the authors investigated computer aided medical diagnosis systems using AIRS method and they claim that, the proposed approach obtained accuracy of 100% and it was a very promising with respect to other classification application problems. In addition, Lahsasna et al. [75] designed a fuzzy based decision support system for CHD diagnosis that achieved 84.44% for accuracy, 79.2% for sensitivity and 88.7% for specificity.

4

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

Table 1 Publications on CI techniques in the medical field. Classifier type 1. Single methods

Title of paper (reference)

Aim

A noninvasive method for coronary artery diseases diagnosis using a clinically-interpretable fuzzy rule-based system [32]

Designed a computer-based noninvasive coronary heart disease (CAD) diagnosis system with clinically interpretable instructions. FRB was used to formulate based on neuro fuzzy classifier (NFC). Aimed to build a classifier that can be able to handle the issues of specifying the risks for a patient that suffer from disease (cardiovascular) up to next 10 years. An interpretable gene expression classifier (iGEC) with a precise and compressed FRB proposed for microarray data analysis. The main objectives of iGEC was to optimize concurrently. Presented a new three –dimensional segmentation for lesion that works based on the clustering algorithms of FCM.

1.1. Fuzzy logic sets (FS) 1.1.1. Fuzzy rule base (FRB)

Medical diagnosis of cardiovascular diseases using an interval-valued fuzzy rule-based classification system [33]

Interpretable gene expression classifier with an accurate and compact fuzzy rule base for microarray data analysis [34]

1.1.2. Fuzzy C-mean (FCM)

A fuzzy C-means (FCM)-Based Approach for Computerized Segmentation of Breast Lesions in Dynamic Contrast-Enhanced MR Images [35] Segmentation techniques for tissue differentiation in MRI of Ophthalmology using fuzzy clustering algorithms [36] Modified fuzzy C-mean in medical image segmentation [37]

1.2. Genetic algorithm (GA) 1.2.1. Genetic programming (GP)

Prediction of Cancer Class with Majority Voting Genetic Programming Classifier Using Gene Expression Data [38] Breast cancer diagnosis using genetic programming generated feature [39] The use of genetic programming in the analysis of quantitative gene expression profiles for identification of nodal status in bladder Cancer [40]

1.3. Particle swarm optimization (PSO)

Robust medical image segmentation using particle swarm optimization aided level set based global fitting energy active contour approach [41] Prediction of Parkinson’s disease tremor onset using a radial basis function neural network based on particle swarm optimization [42]

Human Tremor Analysis Using Particle Swarm [43]

1.4. Artificial neural network (ANN) 1.4.1. Multi-layer perceptron (MLP)

Multilayer perceptron tumor diagnosis based on chromatography analysis of urinary nucleosides [44] A multilayer perceptron-based medical decision support system for heart disease diagnosis [45] A Multilayer Perceptron Network for the Diagnosis of Low Back Pain [46]

Presented magnetic resonance imaging segmentation methods to distinguish tissues in Ophthalmology based on normal and abnormal tissues by utilizing FCM algorithms. Suggested a method to get the image clusters automatically. Then, an improved classification of FCM applied to deliver a fuzzy partition. Presented a majority voting GP classifier for microarray data classification. A new technique was proposed by utilizing the feature generated of GP to diagnose the breast cancer. Developed a detection approach for nodal metastasis from molecular profiles of primary urothelial carcinoma tissues. Samples were run through the GP which is working by N-fold cross validation method to produce classifier instructions of limited complexity. The proposed method works based on Chan and Vese algorithm to attain acceptable segmentation performance, regardless of the early choice of the contour. Aimed to determine the possibility of using a radial basis function neural network based on PSO that can be able to identify that Parkinsonian tremors are occurring from local field potential signals. Presented an approach for human tremor analysis by applying PSO. This study addressed Parkinson’s disease and essential tremor. To diagnose the tumor by using MLP as a practical pattern recognition instrument to recognize cancer patients from health. To develop the MLP method to assist the diagnosis of heart disease based on decision support system. To diagnose the low back pain and sciatic by using the training of MLP network. (continued on next page)

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

5

Table 1 (continued) Classifier type

Title of paper (reference)

Aim

1.4.2. Self-organizing map (SOM)

Self-organizing map for cluster analysis of a breast cancer database [47]

Aimed to recognize and describe clusters in a various breast cancer computer aided diagnosis database. Proposed a new approach of SOM and neural-network algorithm to effectively, broadly and concurrently examine the usage of in 60,0 0 0 genes from 29 bacterial species.

Analysis of codon usage diversity of bacterial genes with a self-organizing map (SOM): characterization of horizontally transferred genes with emphasis on the E. coli O157 genome [48] Breast cancer diagnosis using self-organizing map for sonography [49] 1.4.3. Deep learning

Discrimination of breast cancer with microcalcifications on mammography by deep learning [50] Multi-instance deep learning: discover discriminative local anatomies for body part recognition [51] Hierarchical feature representation and multimodal fusion with deep learning for AD/MCI diagnosis [52]

1.4.4. Extreme learning machine (ELM)

1.4.5. Convolutional neural network (CNN)

A deep learning architecture for image representation, visual interpretability and automated basal-cell carcinoma cancer detection [53] Brain MRI morphological patterns extraction tool based on extreme learning machine and majority vote extraction [54]

Using blood indexes to predict overweight statuses: an extreme learning machine based approach [55] A computer aided diagnosis system for thyroid disease using extreme learning machine [56] Automatic detection of cerebral microbleeds from MR images via 3D convolutional neural networks [57] Lung pattern classification for interstitial lung diseases using a deep convolutional neural network [58] Prediction response to neoadjuvant chemotherapy with PET imaging using convolutional neural networks [59]

1.5. Kernel method 1.5.1. Support vector machine (SVM)

Metabolic changes in rat urine after acute paraquat poisoning and discriminated by support vector machine [60] A support vector machine tool for adaptive Tomotherapy treatments: Prediction of head and neck patients criticalities [61]

Aimed to assess the classification of malignant and benignant sonographic breast lesion using neural network and SOM. To enhance the diagnostic accuracy of microcalcifications by evaluating the performance of deep learning method on large dataset for its discrimination. To solve the issue of body pat recognition by proposing a novel framework of deep learning. To propose a new approach for a high level latent and shared feature representation from neuroimaging modalities by using deep learning method Alzheimer disease and Mild Cognitive Impairment. Presented and evaluated the architecture of DL for automated basal cell carcinoma cancer detection Proposed a capable tool to extract regions from brain magnetic resonance image that discriminate healthy controls form subject with probable dementia of the Alzheimer type. By using ELM approach to choose the most discriminant regions and to make the final classification according to majority vote decision based strategy. Aimed to predict overweight statuses (including 297 females and 179 males) by using the ELM method. To diagnose the thyroid disease with computer aided (CAD) system by using ELM approach. A new automatic approach proposed to detect cerebral microbleeds (CMBs) from magnetic resonance (MR) images by using 3D CNN. A CNN approach designed for classification of interstitial lung diseases (ILDs). To predict neoadjuvant chemotherapy using CNN approach.

Developed a urine metabonomic approach for analyzing the outcome of acute paraquat poisoning on rats and SVM was is used to distinguish the metabolic changes of paraquat groups. Aimed to compound the variables into an anticipating model, to classify criticalities of particular parts and to limit re-planning to designated patients, then letting adaptive radiation therapy methods to be supportable in terms of medical exercise for appropriately designated patients. (continued on next page)

6

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

Table 1 (continued) Classifier type

Title of paper (reference)

Aim

Development of a cervical cancer progress prediction tool for human papillomavirus-positive Koreans: A support vector machine-based approach [62]

To develop a Web-based tool to draw attention to patients positive for human papillomavirus (HPV) who have a high risk of progression to cervical cancer, in order to increase compliance with follow-up examinations and facilitate good doctor–patient communication. Aimed to assess the possibility of utilizing thermal imaging as a possible instrument for recognizing breast cancer by utilizing SVM classifier for automatic classification of normal and malignant breast circumstances. Aimed to predict the status of methylation of CGIs in the human brain consistent with their local genomic sequences by developing a SVM method called MethCGI. A novel approach to analyze the data of DNA microarray experiments which generating thousands gene expression that can be used to collect information for diagnosing the disease was developed by using SVM method.

Thermography based breast cancer detection using texture features and support vector machine [63]

Predicting methylation status of CpG islands in the human brain [64]

Support vector machine classification and validation of cancer tissue samples using microarray expression data [65]

1.6. Artificial immune system (AIS) 1.6.1. Artificial immune recognition system (AIRS)

2. Hybrid methods

2.1. Neuro-fuzzy (NF)

An application of artificial immune recognition System for prediction of diabetes following gestational diabetes [66] Computer aided medical diagnosis system based on principal component analysis and artificial immune recognition system classifier algorithm [67] Prediction of hepatitis disease based on principal component analysis and artificial immune recognition system [68] Prediction of pathological stage in patients with prostate cancer: a neuro-fuzzy model [69] Application of a two-stage fuzzy neural network to a prostate cancer prognosis system. Artificial Intelligence in Medicine [70] Neuro-fuzzy based glucose prediction model for patients with type 1 diabetes mellitus [71]

2.2. Fuzzy support vector machine (FSVM)

Obtaining interpretable fuzzy classification rules from medical data[72] New Fuzzy Support Vector Machine for the Class Imbalance Problem in Medical Datasets Classification [31]

Medical decision support system for diagnosis of neuromuscular disorders using DWT and fuzzy support vector machines [73]

2.3. Fuzzy logic and genetic algorithm (FGA)

Medical image classification based on fuzzy support vector machines [74] Design of a Fuzzy-based Decision Support System for Coronary Heart Disease Diagnosis [75]

To examine the possibility of AIRS method to forecast the gestational diabetes mellitus. Conducted the analysis of lung cancer by computer aided medical diagnosis system based on AIRS and principal component analysis. Proposed the novel approach which used for hepatitis disease diagnosis issue as a classifier. AIRS was used to classify the normalized values input. A neuro-fuzzy model was developed to classify and predict for patients that having organ-confined disease (OCD) or extra-prostatic disease (ED). Two step fuzzy neural network was developed for predictions of prostate cancer. Presented the design and development of a personalized glucose prediction model for patients with type 1 diabetes mellitus and evaluating the proposed design. Discussed extensions to the learning algorithms of NF classification for data analysis. Presented FSVM method for the class imbalance issues which considered class of FSVM by extending manifold regularization and applying two misclassification costs for two classes. Different categories of machine learning techniques were used to categorize electromyography (EMG) signals and compare the related accuracy in classification of EMG signals. Presented a new approach for medical image classification by applying FSVM. Designed a FRB scheme to aid as a decision support system for Coronary heart disease (CHD) analysis that not only considers the decision accuracy of the rules but also their transparency at the same time. (continued on next page)

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

7

Table 1 (continued) Classifier type

Title of paper (reference)

Aim

A fuzzy guided genetic algorithm for operon prediction [76]

Aimed to predict the population of putative operon maps of the genome by developing GA. It used fuzzy functions to assess the fitness to guide their evolution. Presented a GA method using fuzzy logic for computer aided diagnosis in medical imaging. Aimed to focus on the Wisconsin breast cancer diagnosis (WBCD) issue and combined fuzzy system with genetic algorithm (fuzzy-genetic approach). The immune system’s features of learning and memory was presented to eliminate the issues of diagnosis for liver disease. Discussed different studies on chronic obstructive pulmonary and pneumonia disease diagnosis using AIS-NN method.

Medical image classification using genetic-algorithm based fuzzy-logic approach [77] A fuzzy-genetic approach to breast cancer diagnosis [78]

2.4. Artificial immune system and genetic algorithm (AIS-GA)

An Automated Diagnosis System of Liver Disease using Artificial Immune and Genetic Algorithms [79]

2.5. Artificial immune system and neural networks (AIS-NN)

A Comparative Study on Chronic Obstructive Pulmonary and Pneumonia Diseases Diagnosis using Neural Networks and Artificial Immune System [80] Rule extraction from trained adaptive neural network using artificial immune systems [81] Gene selection using hybrid particle swarm optimization and genetic algorithm [82] Hybrid of artificial immune system and particle swarm optimization-based support vector machine for Radio Frequency Identification-based positioning system [83] A hybrid multiclass classifier based on artificial immune algorithm and support vector machine [84] Designing an artificial immune system-based machine learning classifier for medical diagnosis [85] A Support Vector Machine-Firefly Algorithm based forecasting model to determine malaria transmission [86] Intelligent Medical Disease Diagnosis Using Improved Hybrid Genetic Algorithm - Multilayer Perceptron Network [87]

2.6. Genetic algorithm and particle swarm optimization (GA-PSO) 2.7. Artificial immune system and support vector machine (AIS-SVM)

2.8. Support vector machine and firefly algorithm (SVM-FFA) 2.9. Genetic algorithm and multilayer perceptron (GA-MLP)

2.10. Support vector machine and wavelet transform (SVM-WT)

2.11. Support vector machine and particle swarm optimization (SVM-PSO)

An automatic diabetes diagnosis system based on LDA-Wavelet Support Vector Machine Classifier [88]

A prediction approach for multichannel EEG signals modeling using local wavelet SVM [89] A wavelet packet based pulse waveform analysis for cholecystitis and nephrotic syndrome diagnosis[90] PSO-Based Support Vector Machine with Cuckoo Search Technique for Clinical Disease Diagnoses [91] SVM-PSO based Feature Selection for Improving Medical Diagnosis Reliability using Machine Learning Ensembles [92] Liver cancer identification based on PSO-SVM model [93] A combination of modified particle swarm optimization algorithm and support vector machine for gene selection and tumor classification [94]

A new method that used AIS to extract instructions from trained adaptive neural network. The GA-PSO hybrid method was proposed to be used for gene selection. Proposed the AIS-SVM hybrid method based on SVM for enhancing the parameters of SVM that used for radio frequency identification based positioning system. Developed an effective classifier to improve medical diagnosis performance of thyroid gland disease. An efficient method was developed to increase the performance of diagnosis for breast cancer Introduced a novel hybrid approach of SVM-FFA to predict the malaria incidences. This paper introduced an enhanced GA technique which works based on the theory of “most highly fit parents are most likely to produce healthiest offspring”. Introduced a diagnosis method that works automatically for diabetes on linear discriminant analysis (LDA) and Morlet Wavelet Support Vector Machine Classifier. Proposed a novel approach local spatiotemporal prediction using SVM-WT. To explore the possibility of diagnosing cholecystitis and nephrotic syndrome using the pulse waveform data. A new machine learning approach is proposed by utilizing the SVM-PSO method and cuckoo search. Proposed building of rotation forest (RF) ensemble using 20 learners over two clinical datasets namely lymphography and backache. A new technique was used to find the liver cancer by using the hybrid method of PSO-SVM. This paper used the hybrid approach of PSO-SVM for the classification of tumor. (continued on next page)

8

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

Table 1 (continued) Classifier type

Title of paper (reference)

Aim

2.12. Support vector machine and genetic algorithm (SVM-GA)

In silico prediction of mitochondrial toxicity by using GA-CG-SVM approach [95] A robust hybrid between genetic algorithm and support vector machine for extracting an optimal feature gene subset [96]

2.13. Support vector machine and artificial immune recognition system (SVM-AIRS) 2.14. Deep learning and extreme learning machine (DELM)

Diagnosing Tuberculosis With a Novel Support Vector Machine-Based Artificial Immune Recognition System [97] Deep extreme learning machine and its application in EEG classification [98]

Developed a classification mode of SVM mitochondrial toxicity using the GA_CG_SVM scheme. Formalized a robust gene selection method based on a hybrid among GA and SVM to extract fully their respective merits for recognition of main feature genes and for a complex biological phenotype. Introduced a novel approach for diagnosing of tuberculosis using SVM-AIRS method.

2.15. Support vector machine and extreme learning machine (SVM-ELM)

Breast mass classification in digital mammography based extreme learning machine [99]

2.16. Genetic algorithm and fuzzy extreme learning machine

A hybrid automatic system for the diagnosis of lung cancer based on genetic algorithm and fuzzy extreme learning machines [100]. EMG classification in obstructive sleep apnea syndrome and periodic limb movement syndrome patients by using wavelet packet transform and extreme learning machine [101]

2.17. Wavelet packet transform and extreme learning machine (WPT-ELM)

Presented the application of deep leaning and extreme learning machine in EEG classification Proposed a new computer aided diagnosis (CAD) system for breast cancer diagnosis which used the combination of SVM-ELM. Proposed an automatic system using a hybrid approach of GA and fuzzy extreme learning machine to diagnose the lung cancer. Developed a hybrid approach of WPT and ELM to categorize electromyogram (EMG) signals in obstructive sleep apnea syndrome (OSAS) and periodic limb movement syndrome (PLMS) patients.

Table 2 Datasets Classifications. Dataset type

Description

Size

Coronary heart disease (CHD) [32]

This dataset was used from UCI and it is considered as benchmark for different computer aided CAD diagnosis system. This dataset was used with small number of tissue samples and each with expression measurement for thousands of genes. This dataset is used from medical cases of pregnant women in specific medical center located in Taipei, Taiwan. This dataset was obtained from UCI (http://www.ics.uci.edu/∼mlearn/MLRepository.html).

303 records with 76 attributes.

Subset of genes from Ovarian dataset [65] Glucose change test (GCT) [66] Hepatitis disease [68]

Lung cancer [67]

Liver disease [79]

Breast cancer and Echocardiogram[81]

BUPA liver, hepatitis, breast, pima diabetes and heart [31] Leukemia, Colon and breast cancer [82]

ImageCLEF 2007 (X-ray images) [108] TCGA (Prostate cancer) [69]

This dataset was used from the archives of machine learning datasets at University of California, Irvine (http://www.ics.uci.edu./∼mlearn/MLReporsitory.html). The data was used from UCI and it contained of the Indian Liver Patient Dataset [102] . The other dataset was including Liver Disorder dataset from UCI machine learning repository [103] Two different datasets were used as Ljubljana breast cancer and Echocardiogram data from UCI machine learning repository [104]. Five medical datasets from UCI database were used in this research. This study was used 7129 genes and 72 samples for the leukemia dataset [105], the colon dataset includes the expression of 20 0 0 genes in 22 normal tissues and 40 colon tumor tissues [106]. The breast cancer dataset includes 7129 genes and 38 samples [107].

The ImageCLES 2007 [109] medical database was including 116 categories of medical X-ray images. This study used TCGA dataset used the records based on patients diagnosed during the years 20 0 0–2013.

97,802 cDNAs for each tissue. N/A 155 samples, 19 attributes, 13 binary and 6 attributes with 6 to 8 discrete values. 57 attributes.

10 attributes, 583 sample data, and 1 selector field. N/A

N/A 7129 genes and 72 samples for the leukemia dataset. 20 0 0 genes in 22 normal tissues and 40 colon tumor tissues. 7129 genes and 38 samples for breast cancer. 11,0 0 0 medical X-ray images Data were collected from 399 patients

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

9

Table 3 Evaluation criteria (accuracy, sensitivity and specificity) used for CI applied to medical data. Accuracy

Sensitivity

Specificity

Description

T P+T N T P+T N+F P+F N

TP T P+F N

TN T N+F P

N/A

N/A

TP: True positive TN: True negative FP: False positive FN: False negative n: size of dataset D

1 n

 (xi,yi )∈D

(T ) =

δ (I (Di, xi(, yi)

|T

xi : the instance of D yi : the label of xi Di: the possible label of xi by the classifier 1 if i = j δ (i, j ) = { 0 otherwise

i=1

assess(t i ) |T |

N/A

ti ∈ T

N/A

1 i f classi f y (t ) = t.c 0 otherwise T: set of data items to be classified (the test set) Classify(t): classification t by AIRS ti ∈ T & t.c: class of the item t

assess(t )= {

CVA =

1 k

k 

Ai

N/A

N/A

CVA: Cross-validation accuracy

j=1

) 100 × ( Nc Nt

N/A

k: the number of folds used Ai : accuracy measure of each fold, i = 1,…, k Nc : total number of test examples correctly classified by the network Nt : total number examples in test dataset

N/A

Fig. 1. Learning system model.

5. State-of-the-art of single and hybrid CI approaches in medical data 5.1. Single methods In this method, only a single CI classifier (i.e. fuzzy logic sets, GA, PSO, ANN and AIS) has been used to diagnose, analyze and predict diseases for healthcare monitoring system. In addition, CI is used in various procedures to simplify capturing and structuring big data and it also helps examine big data for key insight. Accordingly, the use of AI is increasing in terms of volumes, velocities and variety of data similar to big data [27]. Generally, machine learning is considered as a computational approach by using experience to enhance performance or accuracy of predictions [111]. Fig. 1 illustrates the basic learning model which includes various components such as: i. ii. iii. iv.

Preprocessing phase Learning phase Performance evaluation phase Decision phase

The first phase of intelligent medical diagnosis (IMD) is preprocessing which deals with preparation and transformation of the initial dataset. The medical raw data is transformed into under-

standable format in order to make conceptual labeling. Therefore, the labeled data contains training and testing in the preprocessing phase. The second phase of IMD is learning, which aims to combine a number of learning algorithms into an additional accurate collective class prediction rule. The main rule of this phase is to prepare the input training data for building base classifiers with a learning algorithm as a base learner. Performance evaluation is the third phase of learning algorithm which selects the most appropriate classifier for medical data. In the current research works, the commonly used performance evaluation measures are sensitivity, specificity and overall accuracy. The last step is a decision making process which tunes the learning rules in order to improve the accuracy of the detection. Therefore, it classifies the patterns into normal and abnormal status. 5.1.1. Fuzzy logic sets This method consists of many-valued logic in which the truth values of variables might be any real number between 0 and 1 considered to be fuzzy. It was introduced by Lotfi A. Zadeh who defines the fuzzy sets in mathematics that are sets whose elements have degrees of memberships [112]. The recent activity in healthcare shows the potential use of fuzzy logic sets by using its different approaches as fuzzy rule base (FRB) and fuzzy C-mean (FCM).

10

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

Fig. 2. Tree plan classification of CI techniques in medical data.

5.1.1.1. Fuzzy rule base (FRB). Fuzzy rule-based classification has become well-known in field of medical diagnosis to face the classification problems. It is capable of generating an interpretable model that utilizes the mutual linguistic terms for the user in the problem domain [33,113]. For instance, Marateb and Goudarzi [32] designed a computerbased noninvasive Coronary Artery Diseases/Coronary Heart Disease (CAD/CHDs) diagnosis system with clinically-interpretable rules by using the fuzzy rule-based system. In addition, this study [33] aimed to build a classifier that can be able to handle the issues of specifying the risks for a patient that suffers from a disease (cardiovascular) up to next 10 years. Therefore, it helps doctors in analyzing the efficiency of information given by the system. The other study by Ho et al. [34] introduced an interpretable gene expression classifier (iGEC) with a precise and compressed FRB for microarray data analysis. The main objectives of iGEC was to optimize concurrently, with high classification accuracy, minimal number of rules and minimal number of used genes. The obtained result was 87.9% in terms of accuracy. 5.1.1.2. Fuzzy C-mean (FCM). FCM is considered as unsupervised learning approach in the field of pattern-recognition and it is broadly used in fuzzy clustering algorithm [114]. For instance, Mohamed et al. [37] suggested a method to get the image clusters automatically. Then, an improved classification of FCM is applied to deliver a fuzzy partition. In a different study by Chen et al.

[35] consisting of six consecutive stages for a proposed lesion segmentation algorithm that is applied for clinical MR database containing 121 primary mass lesions and it achieves high accuracy with 97%. In another study [36] the authors used FCM clustering algorithm in segmentation methods to distinguish normal and abnormal tissues in MRI of ophthalmology. It helps decrease the noise effect of medical imaging originating from low resolution sensors or the structures that move during the data acquisition. This can be considered as an assistance to the diagnosis of retinoblastoma and inborn oncological diseases, in which symptoms are typically detected in early childhood in the clinical oncological field. In addition, the recent study by Zhang and Metaxas [115] discusses the large-scale data science methods in medical image analytics and provides numerous benefits for clinical decision making and facilitates efficient medical data management. 5.1.2. Genetic algorithm (GA) GA is fundamentally parallel and it belongs to evolutionary algorithms which are used to produce a beneficial solution for optimization and search problems in the field of AI [116,117]. 5.1.2.1. Genetic programing (GP). GP is an evaluation computation approach and it is a specialization of GA that has been used for machine learning and program induction. It measures how well the computer has performed a task by a set of instructions and fitness

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

11

Table 4 The advantages and disadvantages of single and hybrid methods. Main Types

a) Pros; b) Cons

Methods

Single Method

a) Robustness, scalability and flexibility b) Low accuracy and difficult setting for parameters and metric

Hybrid Method

a) Robustness, flexibility, scalability and adaptability b) Higher accuracy, time consuming in training and testing stage and high resource

Fuzzy logic sets (FRB, FCM) Artificial neural network (SVM, SOM, MLP) Genetic Algorithm (GP) Artificial immune system (AIRS) Particle swarm optimization Neuro-fuzzy Fuzzy support vector machine Genetic Algorithm & fuzzy logic

functions [118]. It also has applications in healthcare for prediction of different diseases. As an example Paul and Iba [38] presented a majority voting GP for the classification of microarray data for prediction of cancer. In other study, Guo and Nandi [39] proposed a new approach to diagnose breast cancer by utilizing the feature generated of GP. Thus, a novel feature extraction measure was developed to overcome the restriction of Fisher principle to support genetic programming optimization features that allow pattern vectors belonging to dissimilar classes to allocate compact and disjoint regions. GP also used for investigation of quantitative expression profiles for recognition of nodal status in bladder cancer in [40] which proposed a detection method for nodal metastasis from molecular profiles of primary urothelial carcinoma tissues. The obtained result shows 81% for accuracy, 60% for sensitivity and 90% for specificity.

5.1.4.1. Multi-layer perceptron (MLP). This method is considered as a feedforward ANN model that maps sets of input data onto a set of suitable outputs. It utilizes a supervised learning approach named backpropagation for training their network which contains several layers of nodes in a directed graph and each layer is linked to the following layer [126]. In 1988, a MLP network was used to diagnose low back pain and sciatic. Then the system performance was compared with three groups of doctors and the other computer program [46]. Another useful example of MLP is one that was established to support the diagnosis of heart diseases [45]. In a similar case [44] the authors applied MLP method to diagnose tumors based on chromatography analysis of urinary nucleoside as a practical pattern recognition instrument to differentiate cancer patients from healthy persons. Their results accuracy was 97% for sensitivity and 85% for specificity.

5.1.3. Particle swarm optimization (PSO) It is a computational approach introduced by Kennedy and Eberhart [119] to optimize an issue by having a population of candidates solution. This work was inspired by social behavior of fish school or bird flock [120,121]. PSO has several similarities with evolutionary computations methods like GA and it is presently a main topic for investigation which provides an alternative to the more recognized evolutionary computation approaches that can be functioning in several of the same domains [122]. PSO is effectively used in various areas like clustering [123] and image watermarking [124]. Besides, healthcare is one of the domains that used the aid of PSO to predict or analyze the diseases and it is also a useful approach for medical images. For instance, the recent research on medical image segmentation used PSO to recommend a robust type of the Chan and Vese algorithm which is anticipated to attain acceptable segmentation performance, regardless of the early choice of the contour [41]. This study used PSO technique to solve the issue of formulating the fitting energy minimization using a metaheuristic optimization algorithm. Eberhart and Hu [43] presented an approach for the examination of human tremor using PSO which addressed two forms of human tremor as important tremor and Parkinson’s disease. PSO is used to progress a neural network that differentiates between normal subjects and those with tremor. Another useful example of PSO which also had the same topic based on previous researches is for prediction of Parkinson’s disease tremor. This method uses a Radial Basis Function Neural Network (RBFNN) based on PSO to detect that Parkinsonian tremors are happening from Local Field Potential (LFP) signals [42].

5.1.4.2. Self-organizing map (SOM). SOM (also called Kohonen map) is an unsupervised learning method of ANN which is capable of prediction, classification, clustering data mining and estimation [127,128]. Several studies have revealed that the great performance of SOM in healthcare and medical fields. For instance, SOM was used in sonography for breast cancer diagnosis to evaluate the classification of malignant and benignant sonographic breast lesions that obtained 85.6% for accuracy, 97.6% for sensitivity and 79.5% for specificity [49]. A year later Kanaya et al. [48] examined how to competently and widely analyze codon usage variety of bacterial genes with SOM. As mentioned above, SOM has great performance in clustering and it was used for cluster analysis to determine clusters in a heterogeneous breast cancer computer aided diagnosis database [47]. Besides, the recent study on breast cancer [129] was developed the different approach as scalable image-retrieval framework for intelligent histopathological image analysis which is evaluated on thousands of histopathological images from breast microscopic tissues. Accordingly, the proposed framework could achieve the classification accuracy of 88.1%.

5.1.4. Artificial neural networks (ANN) ANN is a mature technique in machine learning and cognitive science which is inspired by biological neural networks and it has a wide range of application coverage. It is also used in different applications as image analysis, pattern recognition, adaptive controls and other areas [125].

5.1.4.3. Deep learning (DL). Deep learning has emerged as a new area of machine learning research since 2006 which using a deep graph with multiple processing layers, composed of multiple linear and non-linear transformations [130–133]. In the medical field there are different studies that used deep learning methods for predictions and diagnosis of diseases. It is very useful when applied to large training sets but in the medical field large datasets are not always available [134]. For instance, deep learning was used in hierarchical feature representation and multimodal fusion to diagnose Alzheimer disease (AD) and Mild Cognitive Impairment (MCI) [52]. They evaluated in three binary classifications and the highest obtained result was for AD vs health normal control (NC) with accuracy of 95.35%, sensitivity of 94.65% and specificity of 95.22%. In another study by Wang et al. [50] deep learning was used in breast cancer to enhance the diagnostic accuracy of micro-

12

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

Table 5 Performance evaluation summary for the surveyed CI methods. Method

Reference

Accuracy (%)

Sensitivity (%)

Specificity (%)

Single method Fuzzy rule base (FRB) Fuzzy rule base (FRB) Fuzzy C-mean (FCM) Genetic programming (GP) Genetic programming (GP) Genetic programming (GP) Multi-layer perceptron (MLP) Multi-layer perceptron (MLP) Support vector machine (SVM) Support vector machine (SVM) Support vector machine (SVM) Support vector machine (SVM) Self-organizing map (SOM) Deep learning (DL) Deep learning (DL)

[32] [34] [35] [38] [39] [40] [44] [45] [60] [62] [63] [64] [49] [50] [52]

79% N/A N/A N/A N/A 60% 97% N/A N/A N/A 85.71% 84.32% 97.6% 93% 94.65%

89% N/A N/A N/A N/A 90% 85% N/A N/A N/A 90.48% 84.65% 79.5% 82% 95.22%

Deep learning (DL) Extreme learning machine (ELM) Convolutional neural network (CNN) Convolutional neural network (CNN) Convolutional neural network (CNN) Artificial immune recognition system (AIRS) Artificial immune recognition system (AIRS)

[53] [55] [57] [58] [59] [67] [68]

84% 87.9% 97% 87.5% 98.94% 81% N/A 90% 100% 74.41% 88.10% 84.52% 85.6% 87.3% 95.35% (Alzheimer disease vs healthy normal control) 91.4% 90.54% N/A 85.5% 73.4% 100% 94.12%

88.7% 83.54% 93.16% N/A 80.7% N/A 100%

94.1% 96.80% N/A N/A 81.6% N/A 94.44%

78.9% 96.68% 87. % 78.26% 86% 84.09% 98.25% (Myopathic) 99.5% (Neurogenic) 94.5% (Myopathic) 95.75% (Neurogenic) 79.2% 91.7% 100%

27.4% 96.69% 82.31% 50.0% 69.34% 57.96% 92.25%

90.1%

85.4%

Hybrid methods Neuro-fuzzy (NF) Fuzzy support vector machine (FSVM)

[69] [31]

Fuzzy support vector machine (FSVM)

[73]

81.2% 96.76% (Breast dataset) 83.53% (Heart dataset) 64.44% (Hepatitis dataset) 73.8% (Bupa dataset) 61.94% (Diabetes dataset) 97.67% (Internal cross validation) 93.5% (External cross validation)

90.25%

Fuzzy logic and genetic algorithm (FGA) Fuzzy logic and genetic algorithm (FGA)

[75] [77]

Artificial immune system and genetic algorithm (AIS-GA)

[79]

84.44% 96% (Myocardial heart disease) 88.5% (Microcalcification on mammograms) 88.7% (Dataset of Liver Disorder)

Artificial immune system and neural networks (AIS-NN)

[81]

98.1% (Liver Patient dataset) 94.59% (ECG dataset)

98.9% 96%

96% 91.7%

Genetic algorithm and particle swarm optimization (GA-PSO)

[82]

92.31% (Breast Cancer dataset) 95.1% (Leukemia dataset)

97% N/A

81.18% N/A

[83]

88.7% (Colon dataset) 93.4% (Breast Cancer dataset) 99.47%

N/A

N/A

[84]

99.87%

N/A

N/A

[85]

99.14%

N/A

N/A

[87]

[88]

99.5% (Cancer dataset) 80.4% (Diabetes dataset) 86.3% (Heart dataset) 89.74%

99.2% 79.6% 84.5% 83.33%

100% 83.5% 88.2% 93.75%

[90]

86.57% (Normal Nephrotic syndrome)

N/A

N/A

Support vector machine and particle swarm optimization (SVM-PSO)

[91]

84.82% (Cholecystitis Nephrotic syndrome) 85% (Heart disease dataset)

N/A

N/A

Support vector machine and particle swarm optimization (SVM-PSO)

[92]

91.33% (Breast cancer dataset) 83.72% (Lymphography dataset)

N/A

N/A

Support vector machine and particle swarm optimization (SVM-PSO) Support vector machine and particle swarm optimization (SVM-PSO)

[93]

85.77% (Backache dataset) 76.70%

N/A

N/A

[94]

91.70%

95%

Artificial immune system and support vector machine (AIS-SVM) Artificial immune system and support vector machine (AIS-SVM) Artificial immune system and support vector machine (AIS-SVM) Genetic algorithm and multi-layer perceptron (GA-MLP)

Support vector machine and wavelet transform (SVM-WT) Support vector machine and wavelet transform (SVM-WT)

88.7% 100% 76.9%

91% (continued on next page)

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

13

Table 5 (continued) Method

Reference

Accuracy (%)

Sensitivity (%)

Specificity (%)

Support vector machine and genetic algorithm (SVM-GA) Support vector machine and genetic algorithm (SVM-GA) Support vector machine and artificial immune recognition system (SVM-AIRS) Deep learning and extreme learning machine (DELM) Support vector machine and extreme learning machine (SVM-ELM) Genetic algorithm and fuzzy extreme learning machine Wavelet packet transform and extreme learning machine (WPT-ELM)

[95]

84.59%

88.31%

78.79%

[96]

99%

N/A

N/A

[97]

100%

100%

100%

[98] [99]

82.29% (Diabetes dataset) 96.02%

N/A 96.29%

N/A 94.32%

[100] [101]

98.85% 96.85%

N/A 95.40%

N/A 98.29%

calcifications. The deep learning performance on large datasets was reported to have achieved a discriminative accuracy of 87.3%, sensitivity of 93% and specificity of 82%. Another study of automated basal-cell carcinoma (BCC) cancer detection by Cruz-Roa et al. [53] presented a new method for learning image representation, visual interpretation and automatic BCC cancer detection by using the extension of deep learning architecture to perform digital staining of the input images. The result showed the balanced accuracy of 91.4%, sensitivity of 88.7% and specificity of 94.1%. There are also other successful applications of DL that include Alzheimer’s disease neuroimaging [135], breast density segmentation and mammographic risk scoring [136] and prediction of aqueous solubility for drug like molecules [137]. Recently, a new multi stage DL framework is proposed to solve the issue of body part recognition. The novelty of their work automatically exploits the local information through convolutional neural network (CNN) [138–140] and discovers the discriminative and non-informative local patches via multi instance learning. The proposed approach achieved the 91.67% of accuracy. Additionally, a recent study [141] presented a comprehensive review on digital pathology and microscopy images providing major categories of detection and segmentation methods. The study discussed the existing challenges such as the major issues for automated nucleus/cell detection and segmentation for accurately spate touching or overlapping nuclei/cells. 5.1.4.4. Extreme learning machine (ELM). ELM, proposed by Huang et al. [142], provides a simple and efficient learning algorithm for single hidden layer feedforward neural networks (SFLNs). It has extremely fast learning speed. The hidden nodes are randomly started and then they are fixed without iteratively tuning [143]. Therefore, ELM has many advantages to be used in medical or biomedical data which have high dimensional features. There have been different studies that applied by ELM such as thyroid disease diagnosis [56] which used ELM to assist the tasks using the most discriminative new feature set and the optimal parameters that obtained the maximum accuracy of 98.1%. The ELM approach is also used for blood indexes to predict overweight statuses consisting of 251 healthy subjects and 225 overweight (297 females and 179 males). The results showed the differences in blood and biomedical indexes with accuracy of 90.54%, sensitivity of 83.54 and specificity of 96.80%. In a recent study by Termenon et al. [54] the authors aimed to build a brain MRI morphological patterns extraction tool by using ELM to choose the most discriminant regions and to achieve the final classification according to majority vote decision based strategy. 5.1.4.5. Convolutional neural network (CNN). CNN approach has proven to be an effective and powerful tool for a wide-ranging of computer vision tasks [134]. CNN has been applied in different

medical fields within the past few decades such as medical image processing [144], lung nodule detection [145] and detection of microclassification on mammography [146]. Lately, CNN was used in detection of cerebral microbleeds (CMBs) [57] from MR images by proposing a cascaded framework under 3D CNNs that could achieve a sensitivity of 93.16%. In another recent study by Anthimopoulos et al. [58] the authors proposed and evaluated a CNN method for classification of interstitial lung disease (ILD) with classification performance of 85.5%. Ypsilantis et al. [59] proposed a CNN approach to predict neoadjuvant chemotherapy response based on pretherapy 18Ffluorodeoxyglucose positron emission tomography (18 F-FDG PET) images. The proposed method achieved the accuracy of 73.4%, sensitivity of 80.7% and specificity of 81.6%. 5.1.5. Kernel method This method is a powerful class of algorithms for pattern analysis and one the of the best known kernel based method is support vector machine (SVM) [147, 148]. Kernel method is well performed in different medical fields such as classification of medical dataset [149], medical images [150], gene prioritization [151], ovarian tumors [152], clinical data [153] and etc. 5.1.5.1. Support vector machine (SVM). SVM (also called support vector networks) is proposed by Cortes and Vapnik [154] to solve the issues of binary classification in occasion lacking of statistical sample. In machine learning, SVM is effectively used in addressing the classification issues and regression analysis by analyzing data from supervised learning models with associated learning algorithm [155]. SVMs techniques have also been successfully used in the medical field. In [65] the authors developed new methods to analyze microarray expression data of cancer tissue samples by using SVMs which consist of classification of the tissue samples and an exploration of the data for mislabeled or doubtful tissue result. Moreover, SVM had great performance in predicting methylation status of CpG islands in the human brain [64] obtaining results with 84.52% accuracy, 84.32% sensitivity and 84.65% specificity. The effectiveness of the SVM approach based has been exemplified in a report by Kahng et al. [62] to develop a web-based tool that focuses on patients positive for human papillomavirus (HPV) who have a high risk of progression to cervical cancer. Another useful research was used in the SVM tool to predict head and neck patients criticalities for adaptive tomotherapy treatments [61]. In addition, SVM classifier was used to recognize breast cancer for different conditions (normal and malignant) of breast with accuracy of 88.10%, sensitivity of 85.71% and specificity of 90.48%. 5.1.6. Artificial immune system (AIS) AIS appeared in 1990s as a novel computational intelligence method which is an area of research that links the disciplines of immunology, computer science and engineering [156–158].

14

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

5.1.6.1. Artificial immune recognition system (AIRS). AIRS indicates the most broadly applied AIS-based supervised learning approaches which focus on classification problems [159,160]. AIRS has been used effectively for diagnosing and predicting numerous diseases [97]. The study by Polat and Günes¸ [68] proposed a novel approach for medical applications to be used in hepatitis disease diagnosis. It is primarily based on principal component analysis (PCA) and AIRS. Then, AIRS classifier system helped classify the normalized input values for the analysis and the results show a performance of 94.12% for accuracy, 100% for sensitivity and 94.44% for specificity. In addition, they used the same approach (based on AIRS classifier and PCA) for a medical diagnosis system for lung cancer [67]. In another study, AIRS was used for prediction of gestational diabetes mellitus (GDM) to help doctors in diagnosing pregnant women who have premonition of type 2 diabetes [66]. Nevertheless, some single (i.e. non-hybrid) methods performed well in different studies. However, there is still a need to modify and enhance the existing approaches to get the better performances for medical research. 5.2. Hybrid methods These methods are modifications and/or combinations of more than one AI classifier methods, which helps in enhancing and improving the current system to get better results in healthcare monitoring systems. Concerning the large amount of data related to medical databases and healthcare monitoring systems, the hybrid approach of CI is used to perform better than before. It also assists in structuring and analyzing big data. 5.2.1. Neuro-fuzzy (NF) NF refers to combination of ANN and fuzzy logic that was introduced by J. S. R. Jang. It assists in solving complex problems [161– 163]. It combines human like reasoning style of a fuzzy system with the learning and connectionist structure of neural network [164]. In 1999, NF approach was used for data analysis of medical data to solve the classifier problems [72]. In another example, a NF techniques were performed to capture the metabolic behavior that is used to detect Glucose for the patients who have type 1 diabetes mellitus (T1DM) [71]. The recent study used a twostep fuzzy neural network which is developed for predictions of prostate cancer and the neuro-fuzzy approach which was used for data analysis [70]. In the recent study of prostate cancer [69] the authors developed a NF model to classify and predict the patients that have OCD or ED using The Cancer Genome Atlas (TCGA) dataset and the obtained results were 81.2% for accuracy, 78.9% for sensitivity and 27.4% for specificity. 5.2.2. Fuzzy support vector machine (FSVM) FSVM is another hybrid approach that was developed based on the theory of SVM to solve the problem of over-fitting (which enhances SVM in reducing the result of outlier and noise in data points). One of the essential features of FSVM is that it chooses appropriate fuzzy membership for a given problem [165,166]. For instance, FSVM was used as a new approach to resolve the issues of unclassifiable regions in SVM for medical image classification [74]. Also, in another study FSVM was used to classify electromyography (EMG) signals. It used a discrete wavelet transform (DWT) to achieve an improved performance for internal cross validation. In getting better results in recognition, insensitivity to overtraining, being consistent and demonstrating higher reliability, FSVM was superior to other machine learning approaches [73]. In other recent research by Gu et al. [31] presented a FSVM method in medical datasets classification for the issues of class imbalance by ex-

panding manifold regularization and applying two misclassification costs for two classes and systematically evaluated five different datasets as: pima diabetes, hepatitis, breast, BUPA liver and heart. For instance, the breast dataset results obtained 96.76% for accuracy, 96.68% for sensitivity and 96.69% for specificity. 5.2.3. Fuzzy logic and genetic algorithm (FGA) This method is the combination of fuzzy logic and genetic algorithms (GA) that is broadly used in medical field. For instance, the hybrid approach of fuzzy systems with genetic algorithms which is utilized in the Wisconsin breast cancer diagnosis (WBCD). Therefore, evolutionary algorithms (EA) allow the automatic production of fuzzy systems, based on a database of training cases and this hybrid method produced system exhibits two prime characteristics [78]. Another example by Tsai et al. [77] used GA based on fuzzy logic method for computer aided diagnosis system in medical imaging. The system is used to differentiate myocardial heart disease from echocardiographic images and to identify and categorize clustered microcalcifications from mammograms. This hybrid method helped to get diagnosis with accuracy of 96% for myocardial heart disease and accuracy of 88% at 100% sensitivity level for microcalcification on mammograms. Furthermore, the recent research used one type of fuzzy logic sets for CHD diagnosis. The fuzzy rule-based system (FRBS) was designed to help as a decision support system for CHD diagnosis focusing on accuracy and transparency of the rules at the same time. So, a multi-objective GA was applied to enhance the FRBS transparency and accuracy. Findings of this research show that FRBS is capable to recognize the uncertainty cases with accuracy of 84.44%, sensitivity of 79.2% and specificity of 88.7% [75]. 5.2.4. Artificial immune system and genetic algorithm (AIS-GA) This hybrid method used AIS to assist the GA in increasing the number of possible individuals in the population and it is considered as an alternative in tackling constrained optimization problems [167,168]. For example, AIS and GA were applied to solve liver diagnosis disease issues where the system architecture is AIS-based and the learning procedure of the system adopted by GA interferes with the evolution of antibody population [79]. For instance, the liver patient dataset results were 98.1% for accuracy, 98.9% for sensitivity and 96% for specificity. 5.2.5. Artificial immune system and neural networks (AIS-NN) Both immune system and neural networks are biologically inspired methods and they are capable of identifying patterns of interest. Therefore, there are various similarities and differences among neural systems and immune system according to Dasgupta [169]. AIS and NN are extensively used in several applications and the hybridization of AIS and NN is a useful approach particularly in medical field applications. In [80] the authors provided a comparative study of NN (three different structures) and AIS (one structure) that were used in diseases diagnosis. In another study, a novel method was introduced that uses AIS to exploit instructions from trained adaptive neural networks and the results showed an accuracy of 94.59% for ECG and 92.31% for the breast cancer dataset [81]. 5.2.6. Genetic algorithm and particle swarm optimization (GA-PSO) This method is based upon genetic encoding and natural selection and the PSO based upon social swarm behavior. However, GA and PSO methods are population based algorithms that could be successful in solving a variety of difficult problems and it is possible to hybridize the two methods [170,171]. For instance, GA and PSO were proposed for gene selection and were tested on three benchmark gene expression datasets including breast cancer, colon and leukemia. The effectiveness of this hybrid method shows that it is able to decrease the dimensionality of

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

15

Fig. 3. Comparison of single methods in terms of accuracy, sensitivity and specificity.

the dataset and improve the most informative gene subset and enhance classification accuracy yielding 93.4% accuracy for the breast cancer, 95.1% for the leukemia and 88.7% for the colon dataset [82]. In another example, PSO and GA were used to optimize feature selection using facial and clothing information for gender classification. The PSO-GA were used to choose the best significant features set which more evidently demonstrates the gender and therefore, the data size dimensions can be decreased [172]. 5.2.7. Artificial immune system and support vector machine (AIS-SVM) The hybrid method of AIS and SVM has been used to improve medical diagnosis performance. For example, the combined AIS and SVM were used as classifier for breast cancer where the AIS was accountable for cloning, clustering and mutation of disease patterns and the SVM focused on recognizing the different disease patterns quickly and precisely [85]. In the other similar study [84] the authors used AIS based SVM multiclass classifier to improve the performance of medical diagnosis of thyroid gland disease. The result was 99.87% for accuracy. 5.2.8. Support vector machine and firefly algorithm (SVM-FFA) FFA is proposed by Yang [173] which is a biologically inspired metaheuristic optimization approach and it works based upon certain behavioral patterns, especially the flashing characteristic of fireflies [174]. The combination hybrid methods of SVM and FFA have been used in different applications such as solar radiation [175], lens system [176], stock price forecasting [177], short term load forecasting [178] and other applications. In addition, SVM-FFA hybrid method has been used as forecasting model based in the medical field to determine malaria transmission. This is important for health establishments in order to decide the suitable action for the control of an outbreak. A study by Ch et al. [86] has been applied the FFA for assigning the parameters of SVM in forecasting the malaria occurrences in the Jodhpur and Bikaner area where the transmission of malaria is uncontrolled. 5.2.9. Genetic algorithm and multilayer perceptron (GA-MLP) This hybrid approach was used for intelligent medical disease diagnosis which improves the MLP structure and enhances the accuracy of the medical diagnostic classification. GA was used to automatically examine and concurrently improve the number of hid-

den nodes, initial weight and feature subsets of MLP over the evolution procedure [87]. For example, for the diabetes dataset results were 80.4% for accuracy, 79.6% for sensitivity and 83.5% for specificity. 5.2.10. Support vector machine-wavelet transform (SVM-WT). Wavelet transform (WT) illustrates a mathematical expression for decomposing a time series data into numerous groups in order to get improved analysis of the components [179,180]. Recently, SVM-WT has been broadly used in engineering applications [181–183]. In 2008, there was an investigation on the effect of cholecystitis and nephrotic syndrome on the pulse waveform. So, the pulse waveform signals were decomposed into a given level by wavelet packet transform and the SVM classifiers were trained to decrease the negative effects due to small data sets [90]. In addition, wavelet SVM was proposed as a forecast method for multichannel EEG signal modeling that integrates the local prediction and the wavelet kernel [89]. In other investigation [88] the authors presented an automatic diagnosis system for diabetes based on Morlet wavelet support machine (MWSVM) classifier and linear discriminant analysis (LDA). This research had 3 stages consists of (i) feature extraction, (ii) feature reduction stage by using the LDA method, and (iii) classification stage by using MWSVM classifier. The obtained result was 89.74% for accuracy, 83.33% for sensitivity and 93.75% for specificity. 5.2.11. Support vector machine and particle swarm optimization (SVM-PSO) This hybrid approach is widely applied in the medical field and it has shown significant performances. For example, PSO and SVM were combined as a hybrid method for gene selection and tumor classification where the PSO was used to choose a gene, whereas SVM was use as the classifier. Then, the suggested method was applied to microarray data (22 normal and 40 colon tumor tissues) and it had a great prediction performance of 91.70% for accuracy, 95% for sensitivity and 91% for specificity [94]. In other study, Jiang et al. [93] used a new approach for identifying liver cancer that works based upon PSO-SVM method. Therefore, PSO was applied to select the parameters automatically for SVM. This is an

16

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

Fig. 4. Comparison of hybrid methods in terms of accuracy, sensitivity and specificity.

essential advantage which makes possible the selection of parameter more objectively and it keeps away the subjectivity in the old SVM model. In another research, SVM-PSO was proposed as a feature selection for enhancing medical diagnosis validity by applying machine learning ensembles in order to obtain higher accuracy of ensembles [92]. Liu and Fu [91] introduced a novel method that hybridizes SVM, PSO and Cuckoo Search (CS). This technique uses of 2 steps: (i) developed a method of Cuckoo search for improving the SVM parameters to identify better initial parameter of a kernel function, (ii) PSO is applied to the remainder of the training of SVM and identifies the greatest parameter in SVM. 5.2.12. Support vector machine and genetic algorithm (SVM-GA) This hybrid method also has been used in the medical field. For instance, the hybridization of SVM and GA was able to achieve their corresponding merits in identification of main feature genes for a complex biological phenotype with 99% accuracy [96]. In another example, the SVM approach was combined with GA for feature selection and conjugate gradient (CG) technique for parameter optimization was applied to build a forecasting model of mitochondrial toxicity [95]. 5.2.13. Support vector machine and artificial immune recognition system (SVM-AIRS) This study [97] introduced a new hybrid system of SVM-AIRS for diagnosing tuberculosis with the purpose of increasing the classification accuracy of AIRS. The result shows that the suggested approach was capable of successfully categorizing tuberculosis with 100% accuracy, sensitivity and specificity. 5.2.14. Hybridization of deep learning (DL) method The hybridization of DL also performed well in medical fields. The recent study by Ding et al. [98] focused on the application of hybrid method of deep learning and extreme learning machine (DELM) in EEG classification and they achieved the accuracy of 82.29% for a diabetes dataset. In addition, there are different applications that used DL in medical fields such as chest pathology [184], pathogenicity of genetic variants [185], human genome [186], prediction of lung tumors [187].

5.2.15. Hybridization of extreme learning machine (ELM) method The hybridization of ELM is one of the most effective method in medical fields. For instance, an automatic system was proposed by Daliri [100] to diagnose the lung cancer based on hybrid method of GA and fuzzy extreme learning machine. The proposed system is using the GA for feature selection to discover the features that are more relevant in lung cancer diagnosis. Then, selected features are fed to a fuzzy inference system which is trained using fuzzy extreme learning machines. Thus, the classifier is able to distinguish among different kinds of lung cancers. In another study [101] a hybrid approach of wavelet packet transform (WPT) and ELM was proposed to classify EMG signals in periodic limb movement syndrome (PLMS) and obstructive sleep apnea syndrome (OSAO) patients. Firstly, the WPT was used to extract the features of EMG signals and then these features were fed to ELM classifier. The results showed accuracy of 96.85%, sensitivity of 95.40% and septicity of 98.29%. A recent research in digital mammography [99] the authors proposed a new CAD system based on combination of SVM and ELM methods to diagnose the breast cancer. The proposed CAD system consists of preprocessing, segmentation, feature extraction, feature selection and classification stages. The obtained results were 96.02% for accuracy, 96.29% for sensitivity and 94.32% for specificity. Fig. 2 depicts a tree plan classification of CI techniques in medical data (CIMD). It is classified based on two categories: as single and hybrid methods that are used in medical data for different objectives such as designing, predicting, developing, optimizing and diagnosing in healthcare and medical devices and applications. Table 4 provides the advantages and disadvantages of the healthcare monitoring methods between the single and hybrid methods. As shown in Table 4, hybrid methods have more advantages compare to single method such as adaptability, higher accuracy but they are more time consuming in testing stage. Table 5 provides the experimental data on single and hybrid methods in terms of accuracy, sensitivity, specificity. For instance, AIRS method achieved 100% (accuracy and sensitivity) and 94.44% (specificity) as one of the top computational intelligence method in healthcare and medical fields.

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

Fig. 5. Chronological order of single and hybrid CI techniques applied to medical data.

17

18

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

Hence, the obtained results of single and hybrid methods are shown in Figs. 3 and 4 to get a better understanding by a comparison graph. Fig. 3 provides the comparison of single method in terms of accuracy, sensitivity and specificity. SVM [60], AIRS [67], GP [39], FCM [35] and DL [52] methods had the greatest result in accuracy, AIRS [68], SOM [49], and MLP [44] methods had the highest result in sensitivity. Lastly, ELM [55] and DL [52] methods had the highest result in specificity terms. It shows that SVM, AIRS, DL and ELM methods perform well on different problems where AIRS method is self-adjusting with regard to its construction in problem space [67] and SVM is one of the most successful statistical learning theory algorithm in medicine [60]. Besides that, Fig. 4 compares the hybrid methods which generally had better results compared to single methods. Thus, hybrid methods provide more efficient and accurate results for healthcare monitoring, cancer detection, diseases diagnosis and other medical fields. For instance, the SVM-AIRS, SVM-GA, AIS-SVM and FSVM were the most accurate hybrid methods. In addition, the SVM-AIRS, GAMLP, FSVM, AIS-GA and SVM-ELM were the top hybrid methods in terms of sensitivity. Moreover, the SVM-AIRS, FGA, FSVM, AIS-GA, GA-MLP and WPT-ELM were the top hybrid methods in terms of specificity. It can be seen that the hybridization of SVM with any other methods had great performances that achieved better results in terms of accuracy, sensitivity and specificity. Therefore, SVM is one of the most accurate and robust algorithm, and it is capable to make best use of the predictive accuracy of a model without over fitting the training examples [97]. Fig. 5 provides the chronology of CI approaches that focus on single and hybrid methods which specifies that some approaches originate from other approaches to strengthen their design efficiency and productiveness. For example, in single method division Wen et al. [60] used the SVM approach which is originate from the root of Kernel method (Cortes and Vapnik [154]). In addition, the study of coronary heart disease diagnosis presented by Lahsasna et al. [75] used the hybrid approach of fuzzy-GA which the fuzzy method originate from Zadeh [112] and the GA method originate from [116,117] which elaborated in the chronological graph (Fig. 5).

6. Conclusion A comprehensive taxonomy along with state-of-the-art of single and hybrid CI approaches in healthcare was presented in this paper. This study focused on CI techniques applied to medical data. These complex data are persistently and quickly growing with numerous information values. A comprehensive list of published papers related to CI approaches applied to medical data were reviewed. They were categorized into single and hybrid approaches that were used for prediction, disease diagnosis, cancer detection and healthcare monitoring in medical fields. These approaches were evaluated based on accuracy, sensitivity and specificity. The results of this study indicate that in single methods SVM (100%) and AIRS (100%) had the highest accuracy of detection, SOM (97.6%) and MLP (97%) had the highest sensitivity, and ELM (96.80%) and DL (95.22%) had the highest specificity. In addition, SVM-AIRS method was the best among hybrid methods with 100% accuracy, sensitivity and specificity. The performances of other hybrid methods were GA-MLP (99.5) in accuracy, FGA (100%) and GAMLP (99.2%) in sensitivity and FGA (100%) and GA-MLP (100%) in specificity were the top methods. Our future work is to design a multi agent system based CI applied to medical data with the goal of achieving high performance in detection.

Acknowledgment The authors would like to acknowledge the financial support by University of Malaya Research Grant (UMRG Program, Project number: RP003A-14HNE).

Author Contributions S. Shamshirband and Hamid Alinejad-Rokny designed the study. Ali Kalantari, Amirrudin Kamsin, and Abdullah Gani analyzed the data. Ali Kalantari, Amirrudin Kamsin, S, Shamshirband and Anthony T Chronopoulos analyzed the material and wrote the paper. Hamid Alinejad-Rokny designed the study. References [1] D. Boyd, K. Crawford, Critical questions for big data: provocations for a cultural, technological, and scholarly phenomenon, Inf. Commun. Soc. 15 (2012) 662–679. [2] X. Wu, X. Zhu, G.-Q. Wu, W. Ding, Data mining with big data, IEEE Trans. Knowl. Data Eng. 26 (2014) 97–107. [3] I.A.T. Hashem, I. Yaqoob, N.B. Anuar, S. Mokhtar, A. Gani, S.U. Khan, The rise of “big data” on cloud computing: review and open research issues, Inf. Syst. 47 (2015) 98–115. [4] V. Mayer-Schönberger, K. Cukier, Big Data: A Revolution That will Transform How We Live, Work, and Think, Houghton Mifflin Harcourt, 2013. [5] A. Gani, A. Siddiqa, S. Shamshirband, F. Hanum, A survey on indexing techniques for big data: taxonomy and performance evaluation, Knowl. Inf. Syst. 46 (2016) 241–284. [6] M. Chen, S. Mao, Y. Liu, Big data: a survey, Mobile Netw. Appl. 19 (2014) 171–209. [7] D. Markonis, R. Schaer, I. Eggel, H. Muller, A. Depeursinge, Using MapReduce for large-scale medical image analysis, in: 2012 IEEE Second International Conference on Healthcare Informatics, Imaging and Systems Biology, IEEE, 2012, p. 1. [8] J. Andreu-Perez, C.C. Poon, R.D. Merrifield, S.T. Wong, G.-Z. Yang, Big data for health, IEEE J. Biomed. Health Informatics 19 (2015) 1193–1208. [9] A. Hanbury, H. Müller, G. Langs, M.A. Weber, B.H. Menze, T.S. Fernandez, Bringing the algorithms to the data: cloud-based benchmarking for medical image analysis, in: International Conference of the Cross-Language Evaluation Forum for European Languages, Springer, 2012, pp. 24–29. [10] C.G. Chute, M. Ullman-Cullere, G.M. Wood, S.M. Lin, M. He, J. Pathak, Some experiences and opportunities for big data in translational research, Genet. Med. 15 (2013) 802–809. [11] A. O’Driscoll, J. Daugelaite, R.D. Sleator, ‘Big data’, Hadoop and cloud computing in genomics, J. Biomed. Inform. 46 (2013) 774–781. [12] D.W. Bates, S. Saria, L. Ohno-Machado, A. Shah, G. Escobar, Big data in health care: using analytics to identify and manage high-risk and high-cost patients, Health Aff. 33 (2014) 1123–1131. [13] J.E. Anderson, D.C. Chang, Using electronic health records for surgical quality improvement in the era of big data, JAMA Surg. 150 (2015) 24–29. [14] X.L. Dong, D. Srivastava, Big data integration, in: 2013 IEEE 29th International Conference on Data Engineering (ICDE), IEEE, 2013, pp. 1245–1248. [15] D. Zhang, Inconsistencies in big data, in: 2013 12th IEEE International Conference on Cognitive Informatics & Cognitive Computing (ICCI∗ CC), IEEE, 2013, pp. 61–67. [16] M. Felici, M.G. Jaatun, E. Kosta, N. Wainwright, Bringing accountability to the cloud: addressing emerging threats and legal perspectives, in: Cyber Security and Privacy, Springer, 2013, pp. 28–40. [17] B. Feldman, E.M. Martin, T. Skotnes, Big Data in Healthcare Hype and Hope, 360, Dr. Bonnie, 2012 October 2012. [18] A. Pah, L. Rasmussen-Torvik, S. Goel, P. Greenland, A. Kho, Big data: what is it and what does it mean for cardiovascular research and prevention policy, Curr. Cardiovasc. Risk Rep. 9 (2015) 1–9. [19] K. Kambatla, G. Kollias, V. Kumar, A. Grama, Trends in big data analytics, J. Parallel Distrib. Comput. 74 (2014) 2561–2573. [20] E. Bertino, B.C. Ooi, Y. Yang, R.H. Deng, Privacy and ownership preserving of outsourced medical data, in: ICDE 2005. Proceedings. 21st International Conference on Data Engineering, 2005, IEEE, 2005, pp. 521–532. [21] P.B. Jensen, L.J. Jensen, S. Brunak, Mining electronic health records: towards better research applications and clinical care, Nat. Rev. Genet. 13 (2012) 395–405. [22] K. Shameer, J.C. Denny, K. Ding, H. Jouni, D.R. Crosslin, M. De Andrade, C.G. Chute, P. Peissig, J.A. Pacheco, R. Li, A genome-and phenome-wide association study to identify genetic variants influencing platelet count and volume and their pleiotropic effects, Hum. Genet. 133 (2014) 95–109. [23] B.S. Glicksberg, L. Li, M.A. Badgeley, K. Shameer, R. Kosoy, N.D. Beckmann, N. Pho, J. Hakenberg, M. Ma, K.L. Ayers, Comparative analyses of population-scale phenomic data in electronic medical records reveal race-specific disease networks, Bioinformatics 32 (2016) i101–i110.

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22 [24] J. Sun, C.K. Reddy, Big data analytics for healthcare, in: Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, ACM, 2013 1525–1525. [25] R. Fang, S. Pouyanfar, Y. Yang, S.-C. Chen, S. Iyengar, Computational health informatics in the big data age: a survey, ACM Comput. Surv. (CSUR) 49 (2016) 12. [26] A. Mellit, S.A. Kalogirou, Artificial intelligence techniques for photovoltaic applications: a review, Prog. Energy Combust. Sci. 34 (2008) 574–632. [27] D.E. O’Leary, Artificial intelligence and big data, IEEE Intell. Syst. 28 (2) (2013) 96–99. [28] Y. Koumpouros, Big Data in Healthcare, Healthcare Administration: Concepts, Methodologies, Tools, and Applications: Concepts, Methodologies, Tools and Applications, 2014, p. 23. [29] A.P. Engelbrecht, Computational Intelligence: An Introduction, John Wiley & Sons, 2007. [30] F. Latifogˇ lu, H. Kodaz, S. Kara, S. Günes¸ , Medical application of artificial immune recognition system (AIRS): diagnosis of atherosclerosis from carotid artery Doppler signals, Comput. Biol. Med. 37 (2007) 1092–1099. [31] X. Gu, T. Ni, H. Wang, New fuzzy support vector machine for the class imbalance problem in medical datasets classification, Sci. World J. 2014 (2014) 1–12. [32] H.R. Marateb, S. Goudarzi, A noninvasive method for coronary artery diseases diagnosis using a clinically-interpretable fuzzy rule-based system, J. Res. Med. Sci. 20 (3) (2015) 214. [33] J.A. Sanz, M. Galar, A. Jurio, A. Brugos, M. Pagola, H. Bustince, Medical diagnosis of cardiovascular diseases using an interval-valued fuzzy rule-based classification system, Appl. Soft Comput. 20 (2014) 103–111. [34] S.-Y. Ho, C.-H. Hsieh, H.-M. Chen, H.-L. Huang, Interpretable gene expression classifier with an accurate and compact fuzzy rule base for microarray data analysis, Biosystems 85 (2006) 165–176. [35] W. Chen, M.L. Giger, U. Bick, A fuzzy C-means (FCM)-based approach for computerized segmentation of breast lesions in dynamic contrast-enhanced MR images 1, Acad. Radiol. 13 (2006) 63–72. [36] M.-S. Yang, Y.-J. Hu, K.C.-R. Lin, C.C.-L. Lin, Segmentation techniques for tissue differentiation in MRI of ophthalmology using fuzzy clustering algorithms, Magn. Reson. Imaging 20 (2002) 173–179. [37] N. Mohamed, M. Ahmed, A. Farag, Modified fuzzy C-mean in medical image segmentation, in: Proceedings, 1999 IEEE International Conference on Acoustics, Speech, and Signal Processing, 1999, IEEE, 1999, pp. 3429–3432. [38] T.K. Paul, H. Iba, Prediction of cancer class with majority voting genetic programming classifier using gene expression data, IEEE/ACM Trans. Comput. Biol. Bioinformatics (TCBB) 6 (2009) 353–367. [39] H. Guo, A.K. Nandi, Breast cancer diagnosis using genetic programming generated feature, Pattern Recognit. 39 (2006) 980–987. [40] A.P. Mitra, A.A. Almal, B. George, D.W. Fry, P.F. Lenehan, V. Pagliarulo, R.J. Cote, R.H. Datar, W.P. Worzel, The use of genetic programming in the analysis of quantitative gene expression profiles for identification of nodal status in bladder cancer, BMC Cancer 6 (2006) 159. [41] D. Mandal, A. Chatterjee, M. Maitra, Robust medical image segmentation using particle swarm optimization aided level set based global fitting energy active contour approach, Eng. Appl. Artif. Intell. 35 (2014) 199–214. [42] D. Wu, K. Warwick, Z. Ma, M.N. Gasson, J.G. Burgess, S. Pan, T.Z. Aziz, Prediction of Parkinson’s disease tremor onset using a radial basis function neural network based on particle swarm optimization, Int. J. Neural Syst. 20 (2010) 109–116. [43] R.C. Eberhart, X. Hu, Human tremor analysis using particle swarm optimization, in: 1999. CEC 99. Proceedings of the 1999 Congress on Evolutionary Computation, IEEE, 1999. [44] P. Seidel, A. Seidel, O. Herbarth, Multilayer perceptron tumour diagnosis based on chromatography analysis of urinary nucleosides, Neural Netw. 20 (2007) 646–651. [45] H. Yan, Y. Jiang, J. Zheng, C. Peng, Q. Li, A multilayer perceptron-based medical decision support system for heart disease diagnosis, Expert Syst. Appl. 30 (2006) 272–281. [46] D.G. Bounds, P.J. Lloyd, B. Mathew, G. Waddell, A multilayer perceptron network for the diagnosis of low back pain, in: IEEE International Conference on Neural Networks, 1988, IEEE, 1988, pp. 481–489. [47] M.K. Markey, J.Y. Lo, G.D. Tourassi, C.E. Floyd, Self-organizing map for cluster analysis of a breast cancer database, Artif. Intell. Med. 27 (2003) 113–127. [48] S. Kanaya, M. Kinouchi, T. Abe, Y. Kudo, Y. Yamada, T. Nishi, H. Mori, T. Ikemura, Analysis of codon usage diversity of bacterial genes with a self-organizing map (SOM): characterization of horizontally transferred genes with emphasis on the E. coli O157 genome, Gene 276 (2001) 89–99. [49] D.-R. Chen, R.-F. Chang, Y.-L. Huang, Breast cancer diagnosis using self-organizing map for sonography, Ultrasound Med. Biol. 26 (20 0 0) 405–411. [50] J. Wang, X. Yang, H. Cai, W. Tan, C. Jin, L. Li, Discrimination of breast cancer with microcalcifications on mammography by deep learning, Sci. Rep. 6 (2016) 27327. [51] Z. Yan, Y. Zhan, Z. Peng, S. Liao, Y. Shinagawa, S. Zhang, D.N. Metaxas, X.S. Zhou, Multi-instance deep learning: discover discriminative local anatomies for bodypart recognition, IEEE Trans. Med. Imaging 35 (2016) 1332–1343. [52] H.-I. Suk, S.-W. Lee, D. Shen, A.s.D.N. Initiative, Hierarchical feature representation and multimodal fusion with deep learning for AD/MCI diagnosis, Neuroimage 101 (2014) 569–582.

19

[53] A.A. Cruz-Roa, J.E.A. Ovalle, A. Madabhushi, F.A.G. Osorio, A deep learning architecture for image representation, visual interpretability and automated basal-cell carcinoma cancer detection, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer, 2013, pp. 403–410. [54] M. Termenon, M. Graña, A. Savio, A. Akusok, Y. Miche, K.-M. Björk, A. Lendasse, Brain MRI morphological patterns extraction tool based on extreme learning machine and majority vote classification, Neurocomputing 174 (2016) 344–351. [55] H. Chen, B. Yang, D. Liu, W. Liu, Y. Liu, X. Zhang, L. Hu, Using blood indexes to predict overweight statuses: an extreme learning machine-based approach, PLoS One 10 (2015) e0143003. [56] L.-N. Li, J.-H. Ouyang, H.-L. Chen, D.-Y. Liu, A computer aided diagnosis system for thyroid disease using extreme learning machine, J. Med. Syst. 36 (2012) 3327–3337. [57] Q. Dou, H. Chen, L. Yu, L. Zhao, J. Qin, D. Wang, V.C. Mok, L. Shi, P.-A. Heng, Automatic detection of cerebral microbleeds from MR images via 3D convolutional neural networks, IEEE Trans. Med. Imaging 35 (2016) 1182– 1195. [58] M. Anthimopoulos, S. Christodoulidis, L. Ebner, A. Christe, S. Mougiakakou, Lung pattern classification for interstitial lung diseases using a deep convolutional neural network, IEEE Trans. Med. Imaging 35 (2016) 1207–1216. [59] P.-P. Ypsilantis, M. Siddique, H.-M. Sohn, A. Davies, G. Cook, V. Goh, G. Montana, Predicting response to neoadjuvant chemotherapy with PET imaging using convolutional neural networks, PLoS One 10 (2015) e0137036. [60] C. Wen, Z. Wang, M. Zhang, S. Wang, P. Geng, F. Sun, M. Chen, G. Lin, L. Hu, J. Ma, Metabolic changes in rat urine after acute paraquat poisoning and discriminated by support vector machine, Biomed. Chromatogr. 30 (2016) 75–80. [61] G. Guidi, N. Maffei, C. Vecchi, A. Ciarmatori, G.M. Mistretta, G. Gottardi, B. Meduri, G. Baldazzi, F. Bertoni, T. Costi, A support vector machine tool for adaptive tomotherapy treatments: prediction of head and neck patients criticalities, Phys. Med. 31 (5) (2015) 442–451. [62] J. Kahng, E.H. Kim, H.G. Kim, W. Lee, Development of a cervical cancer progress prediction tool for human papillomavirus-positive Koreans: a support vector machine-based approach, J. Int. Med. Res. 43 (4) (2015) 518–525. [63] U.R. Acharya, E. Ng, J.-H. Tan, S.V. Sree, Thermography based breast cancer detection using texture features and support vector machine, J. Med. Syst. 36 (2012) 1503–1510. [64] F. Fang, S. Fan, X. Zhang, M.Q. Zhang, Predicting methylation status of CpG islands in the human brain, Bioinformatics 22 (2006) 2204–2209. [65] T.S. Furey, N. Cristianini, N. Duffy, D.W. Bednarski, M. Schummer, D. Haussler, Support vector machine classification and validation of cancer tissue samples using microarray expression data, Bioinformatics 16 (20 0 0) 906–914. [66] H.-C. Lin, C.-T. Su, P.-C. Wang, An application of artificial immune recognition system for prediction of diabetes following gestational diabetes, J. Med. Syst. 35 (2011) 283–289. [67] K. Polat, S. Günes¸ , Computer aided medical diagnosis system based on principal component analysis and artificial immune recognition system classifier algorithm, Expert Syst. Appl. 34 (2008) 773–779. [68] K. Polat, S. Günes¸ , Prediction of hepatitis disease based on principal component analysis and artificial immune recognition system, Appl. Math. Comput. 189 (2007) 1282–1291. [69] G. Cosma, G. Acampora, D. Brown, R.C. Rees, M. Khan, A.G. Pockley, Prediction of pathological stage in patients with prostate cancer: a neuro-fuzzy model, PLoS One 11 (2016) e0155856. [70] R.J. Kuo, M.H. Huang, W.C. Cheng, C.C. Lin, Y.H. Wu, Application of a two-stage fuzzy neural network to a prostate cancer prognosis system, Artif. Intell. Med. 63 (2015) 119–133. [71] K. Zarkogianni, K. Mitsis, M.-T. Arredondo, G. Fico, A. Fioravanti, K.S. Nikita, Neuro-fuzzy based glucose prediction model for patients with Type 1 diabetes mellitus, in: 2014 IEEE-EMBS International Conference on Biomedical and Health Informatics (BHI), IEEE, 2014, pp. 252–255. [72] D. Nauck, R. Kruse, Obtaining interpretable fuzzy classification rules from medical data, Artif. Intell. Med. 16 (1999) 149–169. [73] A. Subasi, Medical decision support system for diagnosis of neuromuscular disorders using DWT and fuzzy support vector machines, Comput. Biol. Med. 42 (2012) 806–815. [74] X.-l. Bai, X. Qian, Medical image classification based on fuzzy support vector machines, in: 2008 International Conference on Intelligent Computation Technology and Automation (ICICTA), IEEE, 2008, pp. 145–149. [75] A. Lahsasna, R.N. Ainon, R. Zainuddin, A. Bulgiba, Design of a fuzzy-based decision support system for coronary heart disease diagnosis, J. Med. Syst. 36 (2012) 3293–3306. [76] E. Jacob, R. Sasikumar, K.R. Nair, A fuzzy guided genetic algorithm for operon prediction, Bioinformatics 21 (2005) 1403–1407. [77] D.-Y. Tsai, Y. Lee, M. Sekiya, M. Ohkubo, Medical image classification using genetic-algorithm based fuzzy-logic approach, J. Electron. Imaging 13 (2004) 780–788. [78] C.A. Pena-Reyes, M. Sipper, A fuzzy-genetic approach to breast cancer diagnosis, Artif. Intell. Med. 17 (1999) 131–155. [79] C.L. Liang, L.X. Peng, An automated diagnosis system of liver disease using artificial immune and genetic algorithms, J. Med. Syst. 37 (2013) 10. [80] O. Er, C. Sertkaya, F. Temurtas, A.C. Tanrikulu, A comparative study on chronic obstructive pulmonary and pneumonia diseases diagnosis using neural networks and artificial immune system, J. Med. Syst. 33 (2009) 485–492.

20

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

[81] H. Kahramanli, N. Allahverdi, Rule extraction from trained adaptive neural networks using artificial immune systems, Expert Syst. Appl. 36 (2009) 1513–1522. [82] S. Li, X. Wu, M. Tan, Gene selection using hybrid particle swarm optimization and genetic algorithm, Soft Comput. 12 (2008) 1039–1048. [83] R. Kuo, C. Chen, T.W. Liao, F. Tien, Hybrid of artificial immune system and particle swarm optimization-based support vector machine for radio frequency identification-based positioning system, Comput. Ind. Eng. 64 (2013) 333–341. [84] H.-P. Cheng, C.-S. Cheng, A hybrid multiclass classifier based on artificial immune algorithm and support vector machine, in: 2011 Third International Conference on Data Mining and Intelligent Information Technology Applications (ICMiA), IEEE, 2011, pp. 46–50. [85] H.-P. Cheng, Z.-S. Lin, H.-F. Hsiao, M.-L. Tseng, Designing an artificial immune system-based machine learning classifier for medical diagnosis, in: Information Computing and Applications, Springer, 2010, pp. 333–341. [86] S. Ch, S. Sohani, D. Kumar, A. Malik, B. Chahar, A. Nema, B.K. Panigrahi, R. Dhiman, A support vector machine-firefly algorithm based forecasting model to determine malaria transmission, Neurocomputing 129 (2014) 279–288. [87] F. Ahmad, N.A.M. Isa, Z. Hussain, M.K. Osman, Intelligent medical disease diagnosis using improved hybrid genetic algorithm-multilayer perceptron network, J. Med. Syst. 37 (2013) 1–8. [88] D. Çalis¸ ir, E. Dog˘ antekin, An automatic diabetes diagnosis system based on LDA-wavelet support vector machine classifier, Expert Syst. Appl. 38 (2011) 8311–8315. [89] M. Shen, L. Lin, J. Chen, C.Q. Chang, A prediction approach for multichannel EEG signals modeling using local wavelet SVM, IEEE Trans. Instrum. Measure. 59 (2010) 1485–1492. [90] Q.-L. Guo, K.-Q. Wang, D.-Y. Zhang, N.-M. Li, A wavelet packet based pulse waveform analysis for cholecystitis and nephrotic syndrome diagnosis, in: International Conference on Wavelet Analysis and Pattern Recognition, 2008. ICWAPR’08, IEEE, 2008, pp. 513–517. [91] X. Liu, H. Fu, PSO-based support vector machine with Cuckoo search technique for clinical disease diagnoses, Sci. World J. 2014 (2014) 1–7. [92] I. Mandal, SVM-PSO based Feature Selection for Improving Medical Diagnosis Reliability Using Machine Learning Ensembles 1, 2012. [93] H. Jiang, F. Tang, X. Zhang, Liver cancer identification based on PSO-SVM model, in: 2010 11th International Conference on Control Automation Robotics & Vision (ICARCV), IEEE, 2010, pp. 2519–2523. [94] Q. Shen, W.-M. Shi, W. Kong, B.-X. Ye, A combination of modified particle swarm optimization algorithm and support vector machine for gene selection and tumor classification, Talanta 71 (2007) 1679–1683. [95] H. Zhang, Q.-Y. Chen, M.-L. Xiang, C.-Y. Ma, Q. Huang, S.-Y. Yang, In silico prediction of mitochondrial toxicity by using GA-CG-SVM approach, Toxicol. In Vitro 23 (2009) 134–140. [96] L. Li, W. Jiang, X. Li, K.L. Moser, Z. Guo, L. Du, Q. Wang, E.J. Topol, Q. Wang, S. Rao, A robust hybrid between genetic algorithm and support vector machine for extracting an optimal feature gene subset, Genomics 85 (2005) 16–23. [97] M.R. Saybani, S. Shamshirband, S.G. Hormozi, T.Y. Wah, S. Aghabozorgi, M.A. Pourhoseingholi, T. Olariu, Diagnosing tuberculosis with a novel support vector machine-based artificial immune recognition system, Iran. Red Crescent Med. J. 17 (4) (2015) 1–8. [98] S. Ding, N. Zhang, X. Xu, L. Guo, J. Zhang, Deep extreme learning machine and its application in EEG classification, Math. Probl. Eng. 2015 (2015) 1–11. [99] W. Xie, Y. Li, Y. Ma, Breast mass classification in digital mammography based on extreme learning machine, Neurocomputing 173 (2016) 930–941. [100] M.R. Daliri, A hybrid automatic system for the diagnosis of lung cancer based on genetic algorithm and fuzzy extreme learning machines, J. Med. Syst. 36 (2012) 1001–1005. [101] N. Sezgin, EMG classification in obstructive sleep apnea syndrome and periodic limb movement syndrome patients by using wavelet packet transform and extreme learning machine, Turk. J. Electr. Eng. Comput. Sci. 23 (2015) 873–884. [102] J.A. Blackard, UCI repository of machine learning databases, 2012. http:// archive.ics.uci.edu/ml/datasets/. [103] J.A. Blackard, UCI repository of machine learning databases, 1990. http:// archive.ics.uci.edu/ml/datasets/Liver+Disorders. [104] C. Blake, C.J. Merz, UCI Repository of Machine Learning Databases, 55, University of California, Department of Information and Computer Science, Irvine, CA, 1998 [http://www.ics.uci.edu/∼mlearn/MLRepository.html]. [105] T.R. Golub, D.K. Slonim, P. Tamayo, C. Huard, M. Gaasenbeek, J.P. Mesirov, H. Coller, M.L. Loh, J.R. Downing, M.A. Caligiuri, Molecular classification of cancer: class discovery and class prediction by gene expression monitoring, Science 286 (1999) 531–537. [106] U. Alon, N. Barkai, D.A. Notterman, K. Gish, S. Ybarra, D. Mack, A.J. Levine, Broad patterns of gene expression revealed by clustering analysis of tumor and normal colon tissues probed by oligonucleotide arrays, Proc. Natl. Acad. Sci. 96 (1999) 6745–6750. [107] M. West, C. Blanchette, H. Dressman, E. Huang, S. Ishida, R. Spang, H. Zuzan, J.A. Olson, J.R. Marks, J.R. Nevins, Predicting the clinical status of human breast cancer by using gene expression profiles, Proc. Natl. Acad. Sci. 98 (2001) 11462–11467. [108] M.R. Zare, A. Mueen, W.C. Seng, Automatic classification of medical X-ray images using a bag of visual words, IET Comput. Vis. 7 (2013) 105–114.

[109] H. Müller, T. Deselaers, T.M. Deserno, J. Kalpathy–Cramer, E. Kim, W. Hersh, Overview of the ImageCLEFmed 2007 medical retrieval and medical annotation tasks, in: Workshop of the Cross-Language Evaluation Forum for European Languages, Springer, 2007, pp. 472–491. [110] M.R. Zare, W.C. Seng, A. Mueen, Automatic classification of medical X-ray images, Malaysian J. Comput. Sci. 26 (2013) 9–22. [111] M. Mohri, A. Rostamizadeh, A. Talwalkar, Foundations of Machine Learning, MIT press, 2012. [112] L.A. Zadeh, Fuzzy sets, Inf. Control 8 (1965) 338–353. [113] I. Gadaras, L. Mikhailov, An interpretable fuzzy rule-based classification methodology for medical diagnosis, Artif. Intell. Med. 47 (2009) 25–41. [114] J.C. Bezdek, Pattern Recognition with Fuzzy Objective Function Algorithms, Kluwer Academic Publishers, 1981. [115] S. Zhang, D. Metaxas, Large-scale medical image analytics: recent methodologies, applications and future directions, Med. Image Anal. 33 (2016) 98–101. [116] S.W. Mahfoud, D.E. Goldberg, Parallel recombinative simulated annealing: a genetic algorithm, Parallel Comput. 21 (1995) 1–28. [117] M. Mitchell, An Introduction to Genetic Algorithms, MIT Press, 1998. [118] R. Aler, D. Borrajo, P. Isasi, Using genetic programming to learn and improve control knowledge, Artif. Intell. 141 (2002) 29–56. [119] J. Kennedy, R. Eberhart, Particle swarm optimization, in: 1995. Proceedings, IEEE International Conference on Neural Networks, 1944, 1995, pp. 1942–1948. [120] A.A. Esmin, R.A. Coelho, S. Matwin, A review on particle swarm optimization algorithm and its variants to clustering high-dimensional data, Artif. Intell. Rev. 44 (2015) 23–45. [121] P. Ghamisi, J.A. Benediktsson, Feature selection based on hybridization of genetic algorithm and particle swarm optimization, IEEE Geosci. Remote Sens. Lett. 12 (2015) 309–313. [122] A. Banks, J. Vincent, C. Anyakoha, A review of particle swarm optimization. Part II: hybridisation, combinatorial, multicriteria and constrained optimization, and indicative applications, Nat. Comput. 7 (2008) 109–124. [123] S. Paterlini, T. Krink, Differential evolution and particle swarm optimisation in partitional clustering, Comput. Stat. Data Anal. 50 (2006) 1220–1247. [124] H. Tao, J.M. Zain, M.M. Ahmed, A.N. Abdalla, W. Jing, A wavelet-based particle swarm optimization algorithm for digital image watermarking, Integrated Comput. Aided Eng. 19 (2012) 81–91. [125] C.P. Chen, C.-Y. Zhang, Data-intensive applications, challenges, techniques and technologies: a survey on big data, Inf. Sci. 275 (2014) 314–347. [126] F. Rosenblatt, Principles of Neurodynamics. Perceptrons and the Theory of Brain Mechanisms, 1961 DTIC Document. [127] T. Kohonen, The self-organizing map, Proc. IEEE 78 (1990) 1464–1480. [128] A.M. Kalteh, P. Hjorth, R. Berndtsson, Review of the self-organizing map (SOM) approach in water resources: analysis, modelling and application, Environ. Model. Softw. 23 (2008) 835–845. [129] X. Zhang, W. Liu, M. Dundar, S. Badve, S. Zhang, Towards large-scale histopathological image analysis: hashing-based image retrieval, IEEE Trans. Med. Imaging 34 (2015) 496–506. [130] L. Deng, A tutorial survey of architectures, algorithms, and applications for deep learning, APSIPA Trans. Signal Inf. Process. 3 (2014) e2. [131] G.E. Hinton, S. Osindero, Y.-W. Teh, A fast learning algorithm for deep belief nets, Neural Comput. 18 (2006) 1527–1554. [132] Y. Bengio, Learning deep architectures for AI, Found. Trends® Mach. Learn. 2 (2009) 1–127. [133] Y. Bengio, A. Courville, P. Vincent, Representation learning: a review and new perspectives, IEEE Trans. Pattern Anal. Mach. Intell. 35 (2013) 1798–1828. [134] H. Greenspan, B. van Ginneken, R.M. Summers, Guest editorial deep learning in medical imaging: overview and future promise of an exciting new technique, IEEE Trans. Med. Imaging 35 (2016) 1153–1159. [135] R. Li, W. Zhang, H.-I. Suk, L. Wang, J. Li, D. Shen, S. Ji, Deep learning based imaging data completion for improved brain disease diagnosis, in: International Conference on Medical Image Computing and Computer-Assisted Intervention, Springer, 2014, pp. 305–312. [136] M. Kallenberg, K. Petersen, M. Nielsen, A.Y. Ng, P. Diao, C. Igel, C.M. Vachon, K. Holland, R.R. Winkel, N. Karssemeijer, Unsupervised deep learning applied to breast density segmentation and mammographic risk scoring, IEEE Trans. Med. Imaging 35 (2016) 1322–1331. [137] A. Lusci, G. Pollastri, P. Baldi, Deep architectures and deep learning in chemoinformatics: the prediction of aqueous solubility for drug-like molecules, J. Chem. Inf. Model. 53 (2013) 1563–1575. [138] S. Lawrence, C.L. Giles, A.C. Tsoi, A.D. Back, Face recognition: a convolutional neural-network approach, IEEE Trans. Neural Netw. 8 (1997) 98–113. [139] S. Ji, W. Xu, M. Yang, K. Yu, 3D convolutional neural networks for human action recognition, IEEE Trans. Pattern Anal. Mach. Intell. 35 (2013) 221–231. [140] A. Krizhevsky, I. Sutskever, G.E. Hinton, Imagenet classification with deep convolutional neural networks, in: Advances in Neural Information Processing Systems, 2012, pp. 1097–1105. [141] F. Xing, L. Yang, Robust nucleus/cell detection and segmentation in digital pathology and microscopy images: a comprehensive review, IEEE Rev. Biomed. Eng. 9 (2016) 234–263. [142] G.-B. Huang, Q.-Y. Zhu, C.-K. Siew, Extreme learning machine: theory and applications, Neurocomputing 70 (2006) 489–501. [143] G. Huang, G.-B. Huang, S. Song, K. You, Trends in extreme learning machines: a review, Neural Netw. 61 (2015) 32–48. [144] B. Sahiner, H.-P. Chan, N. Petrick, D. Wei, M.A. Helvie, D.D. Adler, M.M. Goodsitt, Classification of mass and normal breast tissue: a convolution neural

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

[145]

[146]

[147]

[148] [149]

[150]

[151] [152]

[153]

[154] [155]

[156] [157] [158] [159] [160] [161] [162]

[163] [164] [165] [166] [167]

[168]

[169]

[170]

[171]

[172]

[173] [174]

network classifier with spatial domain and texture images, IEEE Trans. Med. Imaging 15 (1996) 598–610. S.-C.B. Lo, J.-S. Lin, M.T. Freedman, S.K. Mun, Computer-assisted diagnosis of lung nodule detection using artificial convoultion neural network, in: Medical Imaging 1993, International Society for Optics and Photonics, 1993, pp. 859–869. H.P. Chan, S.C.B. Lo, B. Sahiner, K.L. Lam, M.A. Helvie, Computer-aided detection of mammographic microcalcifications: pattern recognition with an artificial neural network, Med. Phys. 22 (1995) 1555–1567. A. Daemen, D. Timmerman, T. Van den Bosch, C. Bottomley, E. Kirk, C. Van Holsbeke, L. Valentin, T. Bourne, B. De Moor, Improved modeling of clinical data with kernel methods, Artif. Intell. Med. 54 (2012) 103–114. J. Shawe-Taylor, N. Cristianini, Kernel Methods for Pattern Analysis, Cambridge University Press, 2004. K. Polat, S. Günes¸ , A new feature selection method on classification of medical datasets: kernel F-score feature selection, Expert Syst. Appl. 36 (2009) 10367–10373. D. Gupta, R. Anand, B. Tyagi, A hybrid segmentation method based on Gaussian kernel fuzzy clustering and region based active contour model for ultrasound medical images, Biomed. Signal Process. Control 16 (2015) 98–112. T. De Bie, L.-C. Tranchevent, L.M. Van Oeffelen, Y. Moreau, Kernel-based data fusion for gene prioritization, Bioinformatics 23 (2007) i125–i132. B. Van Calster, D. Timmerman, C. Lu, J.A. Suykens, L. Valentin, C. Van Holsbeke, F. Amant, I. Vergote, S. Van Huffel, Preoperative diagnosis of ovarian tumors using Bayesian kernel-based methods, Ultrasound Obstet. Gynecol. 29 (2007) 496–504. A. Daemen, B. De Moor, Development of a kernel function for clinical data, in: 2009 Annual International Conference of the IEEE Engineering in Medicine and Biology Society, IEEE, 2009, pp. 5913–5917. C. Cortes, V. Vapnik, Support-vector networks, Mach. Learn. 20 (1995) 273–297. C. Sun, Z. Zhang, X. Luo, T. Guo, J. Qu, B. Li, Support vector machine-based Grassmann manifold distance for health monitoring of viscoelastic sandwich structure with material ageing, J. Sound Vib. 368 (2016) 249–263. L.N. De Castro, J. Timmis, Artificial Immune Systems: A New Computational Intelligence Approach, Springer Science & Business Media, 2002. D. Dasgupta, Advances in artificial immune systems, IEEE Comput. Intell. Mag. 1 (2006) 40–49. D. Dasgupta, S. Yu, F. Nino, Recent advances in artificial immune systems: models and applications, Appl. Soft Comput. 11 (2011) 1574–1587. A.B. Watkins, AIRS: A Resource Limited Artificial Immune Classifier, Mississippi State University, 2001. I. Jenhani, Z. Elouedi, Re-visiting the artificial immune recognition system: a survey and an improved version, Artif. Intell. Rev. 42 (2014) 821–833. J.-S.R. Jang, C.-T. Sun, Neuro-fuzzy modeling and control, Proc. IEEE 83 (1995) 378–406. A. Abraham, in: Neuro Fuzzy Systems: State-of-the-Art Modeling Techniques, Connectionist Models of Neurons, Learning Processes, and Artificial Intelligence, Springer, 2001, pp. 269–276. J.-S.R. Jang, ANFIS: adaptive-network-based fuzzy inference system, IEEE Trans. Syst. Man Cybern. 23 (1993) 665–685. S. Kar, S. Das, P.K. Ghosh, Applications of neuro fuzzy systems: a brief review and future outline, Appl. Soft Comput. 15 (2014) 243–259. X. Jiang, Z. Yi, J.C. Lv, Fuzzy SVM with a new fuzzy membership function, Neural Comput. Appl. 15 (2006) 268–276. C.-F. Lin, S.-D. Wang, Fuzzy support vector machines, IEEE Trans. Neural Netw. 13 (2002) 464–471. H.S. Bernardino, H.J. Barbosa, A.C. Lemonge, L. Fonseca, A new hybrid AIS-GA for constrained optimization problems in mechanical engineering, 2008 IEEE Congress on Evolutionary Computation (IEEE World Congress on Computational Intelligence, 2008. H.S. Bernardino, H.J. Barbosa, A.C. Lemonge, Constraint handling in genetic algorithms via artificial immune systems, Late-breaking Paper at Genetic and Evolutionary Computation Conference (GECCO 2006, 2006. D. Dasgupta, Artificial neural networks and artificial immune systems: similarities and differences, in: 1997. Computational Cybernetics and Simulation, 1997 IEEE International Conference on Systems, Man, and Cybernetics, IEEE, 1997, pp. 873–878. Y. Rahmat-Samii, Genetic algorithm (GA) and particle swarm optimization (PSO) in engineering electromagnetics, in: 2003. ICECom 2003. 17th International Conference on Applied Electromagnetics and Communications, IEEE, 2003, pp. 1–5. M. Settles, T. Soule, Breeding swarms: a GA/PSO hybrid, in: Proceedings of the Seventh Annual Conference on Genetic and Evolutionary Computation, ACM, 2005, pp. 161–168. M. Nazir, A. Majid-Mirza, S. Ali-Khan, PSO-GA based optimized feature selection using facial and clothing information for gender classification, J. Appl. Res. Technol. 12 (2014) 145–152. X.-S. Yang, Firefly algorithm, stochastic test functions and design optimisation, Int. J. Bio-Inspired Comput. 2 (2010) 78–84. M.A. Ghorbani, S. Shamshirband, D.Z. Haghi, A. Azani, H. Bonakdari, I. Ebtehaj, Application of firefly algorithm-based support vector machines for prediction of field capacity and permanent wilting point, Soil Tillage Res. 172 (2017) 32–38.

21

[175] L. Olatomiwa, S. Mekhilef, S. Shamshirband, K. Mohammadi, D. Petkovic´ , C. Sudheer, A support vector machine–firefly algorithm-based model for global solar radiation prediction, Sol. Energy 115 (2015) 632–644. [176] S. Shamshirband, D. Petkovic´ , N.T. Pavlovic´ , S. Ch, T.A. Altameem, A. Gani, Support vector machine firefly algorithm based optimization of lens system, Appl. Opt. 54 (2015) 37–45. [177] T. Xiong, Y. Bao, Z. Hu, Multiple-output support vector regression with a firefly algorithm for interval-valued stock price index forecasting, Knowl. Based Syst. 55 (2014) 87–100. [178] A. Kavousi-Fard, H. Samet, F. Marzbani, A new hybrid modified firefly algorithm and support vector regression model for accurate short term load forecasting, Expert Syst. Appl. 41 (2014) 6047–6056. [179] B. Jawerth, W. Sweldens, An overview of wavelet based multiresolution analyses, SIAM Rev. 36 (1994) 377–412. [180] M. Shenify, A.S. Danesh, M. Gocic´ , R.S. Taher, A.W.A. Wahab, A. Gani, S. Shamshirband, D. Petkovic´ , Precipitation estimation using support vector machine with discrete wavelet transform, Water Resour. Manage. 30 (2016) 641–652. [181] Z. Yao, D. Mei, Z. Chen, On-line chatter detection and identification based on wavelet and support vector machine, J. Mater. Process. Technol. 210 (2010) 713–719. [182] Q. Wu, The forecasting model based on wavelet ν -support vector machine, Expert Syst. Appl. 36 (2009) 7604–7610. [183] I. Ebtehaj, H. Bonakdari, S. Shamshirband, K. Mohammadi, A combined support vector machine-wavelet transform model for prediction of sediment transport in sewer, Flow Meas. Instrum. 47 (2016) 19–27. [184] Y. Bar, I. Diamant, L. Wolf, S. Lieberman, E. Konen, H. Greenspan, Chest pathology detection using deep learning with non-medical training, in: 2015 IEEE 12th International Symposium on Biomedical Imaging (ISBI), IEEE, 2015, pp. 294–297. [185] D. Quang, Y. Chen, X. Xie, DANN: a deep learning approach for annotating the pathogenicity of genetic variants, Bioinformatics 31 (5) (2014) 761–763. [186] F. Liu, C. Ren, H. Li, P. Zhou, X. Bo, W. Shu, De novo identification of replication-timing domains in the human genome by deep learning, Bioinformatics 32 (5) (2015) 641–649. [187] S. Park, S.J. Lee, E. Weiss, Y. Motai, Intra-and inter-fractional variation prediction of lung tumors using fuzzy deep learning, IEEE J. Transl. Eng. Health Med. 4 (2016) 1–12. Ali Kalantari received the M.C.S. degrees from the Faculty of Computer Science, University of Malaya (UM). Currently, he is working as research assistant in the Department of Computer Science, UM, Malaysia. His research interests are in Big Data, Cloud Computing and their applications.

Amirrudin Kamsin is a senior lecturer at the Faculty of Computer Science and Information Technology, University of Malaya, Malaysia. He received his B.I.T. (Management) in 2001 and M.Sc. in Computer Animation in 2002 from University of Malaya and Bournemouth University, UK respectively. He obtained his Ph.D. from University College of London (UCL) in 2014. His research areas include human computer interaction (HCI), authentication systems, e-learning, mobile applications, serious game, augmented reality and mobile health services.

Shahaboddin Shamshirband received his Ph.D. in Computer Science from the University of Malaya at Kuala Lumpur, Malaysia in 2014. He is currently a researcher in Faculty of Information Technology, Ton Duc Thang University, Ho Chi Minh City, Vietnam. His areas of research interest are in Big Data, Machine learning, and Network Security. He is a member of IEEE.

22

A. Kalantari et al. / Neurocomputing 276 (2018) 2–22

Abdullah Gani is a professor at the Department of Computer System and Technology, Faculty of Computer Science and Information Technology, University of Malaya, Malaysia. He obtained tertiary academic qualifications from the University of Hull, UK—B.Phil., and M.Sc. (Information Management) and the University of Sheffield, UK for Ph.D. in Computer Science. Prior to his degree studies, he acquired the Teaching Certificate from Kinta Teaching College, Ipoh and Diploma of Computer Science, from ITM. He has vast teaching experience due to having worked in a number of educational institutions locally and abroad—schools, Malay Women Teaching College, Melaka, Ministry of Education; Rotterham College of Technology and Art, Rotterham, UK, University of Sheffield, UK. His interest in research kicked off in 1983 when he was chosen to attend the 3-Month Scientific Research Course in RECSAM by the Ministry of Education, Malaysia. Since then, more than 150 academic papers have been published in proceedings and respectable journals internationally within top 10% ranking. He received a very good number of citation in Web of Science as well as Scopus databases. He actively supervises numerous students at all level of study—Bachelor, Master and Ph.D. Area of interest in research includes Self-Organized Systems, Machine Learning, Reinforcement Learning, Wireless related networks. Hamid Alinejad-Rokny is working in University of Western Australia. His research and development experience includes over 10 years in the Academia. He is the author/co-author of more than 60 publications in technical journals and conferences. He served on the program committees of several national and international conferences. He is Editor-Chief at Journal of Computational Biology, Biotechnology and Machine Learning. He also is editorial board member at IJSEI, IJFIPM and IJSCIP. His research interests are in the areas of data mining, artificial intelligence and bioinformatics and computational biology.

Anthony T. Chronopoulos received his Ph.D. in Computer Science from the University of Illinois at Urbana Champaign, USA in 1987. He is currently a professor in Computer Science at the University of Texas at San Antonio, USA. His areas of research interest are in Distributed Computing, Cloud and Grid Computing, High Performance Computing, Wireless Communications, Networks and Security. He has published 57 journal and 71 peer-reviewed conference proceedings, with over 20 0 0 non-self-citations and h-index = 28. He is a senior member of IEEE (since 1997).