Machine Learning Research Towards CombatingCOVID-19: Virus Detection, Spread Prevention,

and Medical AssistanceOsama Shahid∗, Mohammad Nasajpour∗, Seyedamin Pouriyeh∗, Reza M. Parizi†, Meng Han∗, Maria Valero∗,

Fangyu Li ‡, Mohammed Aledhari§, Quan Z. Sheng¶∗ Department of Information Technology, Kennesaw State University, Marietta, GA, USA{oshahid1, mnasajp1}@students.kennesaw.edu, {spouriye, mhan9, mvalero2}@kennesaw.edu

† Department of Software Engineering and Game Development, Kennesaw State University, Marietta, GA, USArparizi1@kennesaw.edu

‡Department of Electrical and Computer Engineering, Kennesaw State University, Marietta, GA, USAfli6@kennesaw.edu

§Department of Computer Science, Kennesaw State University, Marietta, GA, USAmaledhar@kennesaw.edu

¶Department of Computing, Macquarie University, Sydney, Australiamichael.sheng@mq.edu.au

Abstract—COVID-19 was first discovered in December 2019and has continued to rapidly spread across countries worldwideinfecting thousands and millions of people. The virus is deadly,and people who are suffering from prior illnesses or are olderthan the age of 60 are at a higher risk of mortality. Medicine andHealthcare industries have surged towards finding a cure, anddifferent policies have been amended to mitigate the spread ofthe virus. While Machine Learning (ML) methods have beenwidely used in other domains, there is now a high demandfor ML-aided diagnosis systems for screening, tracking, andpredicting the spread of COVID-19 and finding a cure againstit. In this paper, we present a journey of what role ML hasplayed so far in combating the virus, mainly looking at it froma screening, forecasting, and vaccine perspectives. We present acomprehensive survey of the ML algorithms and models that canbe used on this expedition and aid with battling the virus.

Index Terms—COVID-19, Machine Learning, Artificial Intel-ligence, Healthcare, Drug Development, Prevention, PredictiveAnalysis, Diagnosis, Image Classification.

I. INTRODUCTION

In December 2019, a novel severe contagious respiratorysyndrome coronavirus 2, which is a type of Severe AcuteRespiratory Syndrome (SARS-CoV-2) virus called COVID-19,was discovered in Wuhan, China [1]. COVID-19 virus isairborne and can easily spread and infect people [2]. Ac-cording to the Centers for Disease Control and Prevention(CDC) [3], the infected people show a range of symptomslike dry cough, shortness of breath, fatigue, losing the sense oftaste and smell, diarrhea, and congestion. Infected patients canalso present fever episodes. Strangely enough, some patientswho have contracted the virus might not even show any ofthe aforementioned symptoms [4]. They can feel completely

Corresponding author: Seyedamin Pouriyeh (email: spouriye@kennesaw.edu).

normal carrying the virus and continuing to spread the diseasewithout knowing [4]. As COVID-19 has a rapid nature ofspreading, the World Health Organization (WHO) declaredit as a global pandemic in March 2020 [5]. At the time ofwriting this paper (i.e., September 2020), the total number ofconfirmed COVID-19 cases worldwide was over 32 million[6]. To tackle this outbreak, scientists in different researchcommunities are seeking a wide variety of computer-aidedsystems such as the Internet of Things [7], Machine Learning(ML) or Deep Learning (DL) techniques [8], [9], Big Data[10], and Blockchain [11] that can assist with overcoming thechallenges brought by COVID-19. These technologies can beused for controlling the spread of the virus, detecting the virus,or even designing and manufacturing a vaccine or drug tocombat it.

There were two epidemics in the past from the coronavirusfamily including Severe Acute Respiratory Syndrome (SARS-CoV) [12] and Middle Eastern Respiratory Syndrome (MERS)[13]. SARS-CoV is a respiratory virus that was transmissiblefrom person to person and it was first identified in 2003. Thevirus had over 8,000 confirmed cases worldwide during itscourse which affected over 26 countries [14]. MERS is also arespiratory virus with similar symptoms of SARS-CoV.

ML, as a subset of Artificial Intelligence (AI), has shown alot of potentials in many industries like retail [15], banks [16],healthcare [17], [18], pharmaceuticals [19], and many more[20]. ML techniques can be programmed to imitate humanintelligence. For example, in the healthcare industry, MLtechniques can be trained and used towards medical diagnosis[21]. ML models have been vastly trained over a datasetconsisting of medical images like Computed Tomography (CT)Scan, Magnetic Resonance Imaging (MRI), or X-Ray to detect

arX

iv:2

010.

0703

6v1

[cs

.CY

] 2

9 Se

p 20

20

anomalies [22], [23]. Its classification models can be expandedin diverse areas including cancer [24], diabetes [25], fatty liver[26], etc. As an example, breast cancer can be diagnosed witha prediction accuracy of 97.13% [27] using ML models.

During previous epidemics, ML techniques have beenwidely implemented in order to assist healthcare authoritiesfor better actions regarding the diseases [28]. For example,Sandhu et al. [29] proposed an ML model that utilizes GPStechnology along with cloud computing power and GoogleMaps to represent potentially infected patients and provide analternative route for uninfected users resulting in potentiallymitigating the spread. The model reaches the classificationaccuracy of 80% in re-routing away from infected patients. Inanother study, Choi et al. [30] used ML models for sentimentalanalysis to review public overreaction appearing in mediaarticles and social media platforms. This type of in-depthanalysis can rapidly monitor the public reaction. It can alsoaid policymakers in taking the right actions in reducing fearand distress from the public regarding MERS.

ML has also been widely used in order to improve clinicaldecision-making regarding the current COVID-19 pandemic[31]. Researchers, using ML algorithms and clustering tech-nique, are able to forecast the spread in provinces [32], [33].ML methods of image classification are used by the scientificcommunity to help in diagnosing the deadly virus [34]. Withthe objective of finding a cure for the virus, ML algorithmsare used to evaluate how dependable are off-the-counter drugsmay be used to help infected patients [35].

The aforementioned examples show the potential of ML inthe detection, diagnosis and prediction of viruses. This paperlargely provides a survey reviewing the research that has beendedicated by the scientific community in using ML technologyin combating COVID-19. In particular, we investigate the roleof ML in detecting or screening, forecasting, and medicalassistance for the virus.

The remainder of the paper is organized as follows. In Sec-tion II, we review the role of ML in detecting and the screeningprocess of COVID-19. We study the use of different MLtechniques regarding four major sections for diagnosing andscreening including Medical Imaging, Chatbot, and ArtificialIntelligence of Things (AIoT). In Section III, the importance ofdisease contamination and its exposure to others are discussedregarding the use of ML for tracking and predicting thespread of COVID-19. This section is mainly divided into threeparts reviewing preventing the spread, contact tracing, andforecasting. Similarly, Section IV reviews the need for medicalassistance during the pandemic and how ML technology canbe integrated. Understanding the virus and developing a drugor vaccine using ML techniques are discussed in this sectionas well. Finally, we discuss, outline future work, and concludein sections V and VI respectively.

II. ML TECHNIQUES TOWARDS DETECTING & SCREENINGCOVID-19

Detecting either a symptomatic or asymptomatic disease inan early stage could be highly effective in order to start the

process of treatment. Regarding the COVID-19, not only ithelps to avoid the spread of contamination, but also it is cost-beneficial. The standard method of diagnosing COVID-19 isto conduct Reverse-Transcription Polymerase Chain Reaction(RT-PCR) test [36]. The RT-PCR is a swab test that is usedto detect nucleic acid from COVID-19 in the upper andlower respiratory system. At the beginning of the pandemic,the sensitivity of the RT-PCR test could show negative forpatients who were later in-fact confirmed positive, hence thereporting of false-negative rates was high [37]. There was alsoa concern about having a shortage and a limited number oftests, and the high-cost factor of producing and conductingthem [38]. It is important to explore alternative methodsof diagnosing COVID-19 that would speed-up the process[39]. To help overcome the challenge presented in diagnosingCOVID-19. ML models and algorithms have shown promisingresults in different stages of COVID-19 [40]. In general, MLtechniques have been widely utilized in the healthcare domainand similarly, it can be used towards analyzing data anddiagnosing COVID-19 using medical imaging which includesX-Ray and CT Scan images [34]. In this section, we reviewdifferent ML techniques that have been used for screeningand diagnosis of COVID-19. Moreover, we discuss other ML-based tools including Chatbots and Artificial Intelligence ofThings (AIoT).

A. Medical Imaging

Diagnosing COVID-19 is one of the most important partsof dealing with the disease. As a result of low access andhigh possibility of false-negative results to the RT-PCR kits,there is an essential need for using other approaches such asmedical images analysis for accurate and reliable screeningand diagnosis in COVID-19 [41]. In general, analyzing medi-cal imaging modalities such as chest X-ray and CT-Scan havekey contributions in confirming the diagnosis of COVID-19 aswell as screening the progression of the disease [42]. DifferentML techniques that incorporate X-ray and CT-Scan imageprocessing approaches could help physicians and healthcareprofessionals as a better way for diagnosis and understandingof the progression of the COVID-19 disease.

1) X-ray: During this pandemic, chest imaging can be animportant part of the COVID-19 in early stage of detection.Classifying patients rapidly is what is expected from theseapproaches. Within the categorization of medical imaging,Chest X-Ray (CXR) was recommended to be implemented asthe first medical imaging regarding COVID-19 by the ItalianSociety of Radiology (SIRM) [43]. Figure 1 demonstrates theCXR images from infected and normal people. Accordingto Cozzi et al. [44], CXR has a sensitivity of 67.1% whichcan be first implemented in special cases including assistingradiologists with better COVID-19 cases identification andfast treatment assigning to the patient. Additionally, CXR isinexpensive and secure because of minimizing the risk ofcontamination which makes a safer workplace for healthcareworkers as well.

In order to decrease the amount of work by radiologists, MLtechniques can be assigned to classify patients with respectto COVID-19. To do that, researchers are mostly focused onthe ML classification models such as Support Vector Machine(SVM), Convolutional Neural Networks (CNN), DL. Oneapproach [45] implemented X-Ray in order to classify the lunglesions (caused by COVID-19) with Multi-level Threshold(MT) process and SVM model. Within this model, firstly,the lung image contrast will be enhanced. Secondly, theimage will be reduced into specific sections (using MT) toavoid duplication of work on uninfected areas. Lastly, theSVM model classifies the sections of the lung with respectto the predefined healthy lungs. Sethy et al. [46] developeda platform using a variety of Deep Convolutional NeuralNetworks (DCNN) models classifying within the SVM withtwo different datasets in order to detect COVID-19 cases basedon the related CRX image. Similarly, [47] proposed a DCNNmodel using the data gathered from two hospitals in Italy torepresents the importance of AI in the detection of COVID-19.

Zhang et al. [48] trained ML models over a large viralpneumonia dataset of CXR images to detect anomalies. Theytested their model on a completely different dataset that hasCOVID-19 CXR images. This is done as one of the symptomsof COVID-19 can be pneumonia [3]. The results are impres-sive as the model performs well when tested on the COVID-19dataset with the Area Under the Curve (AUC) of 83.61%. It iseven more impressive as the model was trained on a differentdataset and yet performed well. Similarly, Wang et al. [49]utilized COVIDx dataset, a publicly available dataset consistsof COVID-19, pneumonia and non-COVID-19 pneumonia-related X-ray images. The authors used this data to train theirmodel for detection of COVID-19, the Deep Neural Network(DNN) is referred to as COVID-Net showing promising resultsin diagnosing infected patients. Apostolopoulos et al. [50]used transfer learning approaches like feature extraction andfine-tuning of CNN based models and trained and tested oversimilar datasets achieving a prediction accuracy up to almost98%. They demonstrated that implementing transfer learningcan have a significant improvement in results. Most ML classi-fiers are trained and tested to achieve high prediction accuracyof COVID-19; however, it is also important to quantify theuncertainty that could exist by using such classifiers as aprimary medium of diagnosis. An approach to validate theML prediction of diagnosis in CXR images was reviewed byGhoshal et al. [51]. It exploited a Bayesian Deep Learningclassifier to estimate the model uncertainty. The result analysisdisplays a strong correlation between uncertainty and accuracyof prediction, which means that the higher the uncertaintyoutcome, the more reliable the prediction accuracy.

Many other ML models are utilized for detectingCOVID-19, and a subset of them are presented in Table I.The references are both published papers and also papers thatare yet to be peer-reviewed. All references that are presented inthe table show high performance towards detecting predictingCOVID-19 through CXR images.

2) CT-Scan: Another applicable medical imaging tool forCOVID-19 diagnosis is a chest Computed Tomography (CT)Scan which is more accurate in detecting COVID-19 cases[79]. Due to respiratory problems of COVID-19 which includelung abnormalities, CT-Scan can be specified as the detectingprocedure for the early stage of a pandemic while none ofthe COVID-19 symptoms appear in patients [80]. Figure 2demonstrates the CT-Scan images from infected and normalpeople.

Ardakani et al. [81] implemented a Computer-aided diagno-sis system to show the benefits of DL in diagnosing COVID-19using a variety of CNNs which conclude the ResNet-101 asthe most precise model. The proposed model can lower theworkload of radiology workers as well. Similarly, Li et al. [82]developed a DL framework known as COVID-19 detectionneural network (COVNet) which can differentiate COVID-19from typical types of pneumonia using chest CT-Scan images.An ML model for quantitative infection assessment throughCT Scan images was modeled by Shan et al. [83] suggestingtheir model is capable of estimating the shape, volume, and thepercentage of infection. A Human-In-The-Loop method [84]was proposed that involves healthcare workers to intervenewith the VB-Net (a modified 3D CNN that combines V-Netand bottle-neck structure) ML model to add more CT-Scanimages into the training model to constantly update the modeland produce more efficient results. Tang, et al. [85] introducedanother method to detect the severity of COVID-19 throughassessing quantitative features from CT-Scan images. The useof the Random Forest (RF) model including 500 decisiontrees along with three-fold cross-validation allows authors tocalculate 63 quantitative features of COVID-19 like infectionvolume or lung ratio. However, the model is limited by abinary classification i.e., results can either be severe or non-severe when they should be mild, common, severe, and critical.Gozes et al. [86] trained clinical models integrated with MLto detect the virus that achieves high accuracy and is also ableto quantify the burden of the disease.

Similar to the previous section, ML techniques that canbe used for detecting COVID-19 from CT-Scan images arepresented in Table II. The list of references includes bothpublished and yet to be peer-reviewed.

B. Chatbots

Computer programs developed to communicate with hu-mans by adopting natural languages are called chatbots [103].A chatbot basically can communicate with different usersand generate proper responses to those users based on theirinputs. Recently, the COVID-19 pandemic has led to buildingdifferent chatbots instead of using hotlines as a communicationmethod. This will reduce hospital visits and increase theefficiency of communicating [104]. Generally, chatbots areimplemented in order to provide an online conversation withthe user by either text or voice displays on web applications,smartphone applications, channels, and so on [105]. Thisconversation can help the user to have a better understandingof his or her situation and gives some hints to users so

Fig. 1. Chest X-Ray (CXR) images of COVID-19 infected people versus uninfected people [52].

TABLE IML RESEARCH DONE TOWARDS DIAGNOSING COVID-19 USING X-RAY (CXR) DATASETS.

Reference Dataset Methods Remarks[50] Multiple Datasets that include 448 confirmed COVID-19 images source: Github DL - CNN, Feature Extraction (various models) Various model performance comparison

[51] 68 COVID-19 Chest X-ray images and 5873 Pneumonia images source: Github, Kaggle Bayesian Deep Learning Classifiers Using Transfer learning approach with the classifier to estimate uncertainty

[53]

COVIDx

Fine Tuning - ResNet Model Model achieves high accuracy for multi-class classification[54] Multiple models VGG19, MobileNet Test and train multiple image classifiers to obtain the highest accuracy identifying the virus;

VGG19 and DenseNet have a high accuracy score[55] Multiple Image Classification Models Demonstrate how DNNs are vulnerable to a universal adversarial perturbation causing failure

in classification tasks, fine-tuning may be able to improve this[56] Capsule Network-based framework - COVID-CAPS An alternative modeling framework that has decent performance with low trainable parameters[57] DenseNet-121 The model is explained well, to test its robustness authors perform multi-class classification and perform k-fold validation[58] SqueezeNet-Bayesian based model The authors classify input of X-Ray images of Normal, COVID-19,

Pneumonia using data augmentation and fine-tuning techniques[59] CoroNet - Semi-supervised learning technique based on AutoEncoders[60] EfficientNet The proposed approach is able to produce a model with high quality with high accuracy[49] COVID-Net COVID-Net is publicly accessible for scientists to further build and improve the network to achieve high accuracy[61] CXR images of 50 COVID-19 patients, and 50 Normal CXR images Three different CNN models (InceptionV3, ResNet50, Inception-ResNetV2) ResNet50 provides the highest classification.

[62] 170 Chest X-Ray images of 45 patients from 5 different sources Modified Pre-trained AlexNet and a simple CNN Pre-Trained network achieves a higher accuracy

[63] 295 COVID-19 CXR images and 163 Pneumonia and Normal CXR images MobileNetV2/SqueezeNet They use DL models achieve high classification rate

[64] 127 COVID-19 CXR images and 1000 Pneumonia and Normal CXR images DarkCovidNet (CNN) Provides two approaches, a binary classification and multi-class classification. Binary model performs better and higher

[65] A large dataset containing both COVID-19 and Non COVID-19 cases from multiple sources VGG16, InceptionV3, Xception, DenseNet-121, NasNet-Mobile, etc models compared VGG16 achieved the highest rate

[66] 306 COVID-19 CXR images and 113 normal CXR images CNN model using Decision Tree Classifier The tested method appears to be robust and provides results that are accurate

[67] CXR images of confirmed 150 COVID-19 patients from Wuhan source: Kaggle Convolutional Neural Network The authors are able to get 93% accuracy

[46] Multiple datasets with 183 COVID-19 CXR images and CXR images of SARS-CoV and MERS Trained 9 different models for COVID-19 ResNet50 + SVM achieved the highest accuracy

[68] 231 COVID-19 CXR images, and 2100 Pneumonia and normal CXR images source: Github Novel ANN - Convolutional CapsNet Model provides highly accurate diagnostics with Binary Classification (whether COVID-19 or no finding)

[69] 423 COVID-19 CXR images, 3064 CXR images of normal CXR and viral pneumonia 8 Different CNN Models Transfer Learning and Data augmentation Techniques used, CheXNet performs the best

[70] 192 COVID-19 CXR images, and 145 normal CXR images nCOVnet(VGG-16) The proposed model is able to achieve high accuracy in predicting COVID-19 infected patients from CXR Images

[71] 10 CXRs from COVID-19 confirmed patients in China and USA DL U-Net Model Model shows great promise, with potential use towards early diagnosis for COVID-19 pneumonia

[72] 126 COVID-19 CXR images, 5835 normal and pneumonai CXR images GSA-DenseNet121-COVID-19 (Hybrid CNN using Optimization algorithm) The proposed model achieves high accuracy in diagnosing, up to 98%

[73] 250 COVID-19 CXR images, 4934 Non COVID-19 CXR images ResNet18, ResNet50, SqueezeNet and DenseNet-121 The models perform well and are tested across multiple parametrs such as Receiver Operating Characteristic (ROC), precision-recall curve, etc.

[74] Multiple datasets including 162 COVID-19 positive CXR images, and Non COVID-19 CXR mages Truncated Inception Net The proposed model achieves an accuracy of 99.92% (AUC 0.99) in classifying COVID-19 positive cases

[75] 305 COVID-19 CXR images, and 822 Non COVID-19 CXR images Transfer Learning method employed on pre-trained models The authors use Gradient Class Activation Map for detecting where the model focuses more on for classification

[76] 180 COVID-19 CXR images, and Non COVID-19 CXR images Xception and ResNet50V2 A concatenated of the two models performs well towards detecting COVID-19

[77] 318 COVID-19 CXR images, and Non COVID-19 CXR images COVID-DA Propose a Deep Learning model that has a novel classifier separation scheme

[78] 455 COVID-19 CXR images, and 3450 Non COVID-19 CXR images MobileNetV2 Training the CNN MobileNetV2 model from scratch proves to get higher accuracy compared to transfer learning techniques

that he or she can take proper steps. Chatbots are usuallyconsidered as one of the best suited to screen patients remotelywithout interactions [106]. The advantages of them includequickly updating information, repetitively encouraging newbehaviors such as washing hands, and assisting with psy-chological support due to the stress caused by isolation andmisinformation [107]. The ML-based chatbots are improvedduring the training procedure and using more data makesthis approach more reliable. During the COVID-19 pandemic,chatbots are getting more attention in order to provide moredetails about COVID-19 in different stages. A wide varietyof chatbots with different languages have been implementedto help patients at the early stage of COVID-19. “AapkaChkitsak”, an AI-based chatbot developed by [108] in India,assists patients with remote consultation regarding their healthinformation, and treatments. This application is developed on

Google Cloud Platform with the main assistance of NaturalLanguage Processing (NLP) which is compatible with eitherspeech or text. Similarly, Ouerhani et al. [109] developed achatbot (called “COVID-Chatbot”) based on DL model whichuses NLP in order to enhance the awareness of people aboutthe ongoing pandemic. COVID-Chatbot is implemented todecrease the impact of the disease during and after the quar-antine phase. Another example is Bebot [110] that providesupdated data regarding the pandemic, and also assists patientswith symptoms checking. Some other implemented chatbotsduring this unprecedented time are including Orbita [111],Hyro [112], Apple’s screening tools (website, application,voice command or Siri) [113], CDC’s self-checker [114], andSymptoma [115]. A brief description of these chatbots can befound in Table III.

Fig. 2. CT-Scan images of COVID-19 infected people versus uninfected people [52].

TABLE IIML RESEARCH DONE TOWARDS DIAGNOSING COVID-19 USING CT-SCAN DATASETS.

Reference Dataset Methods Remarks[87] Open-sourced COVID-CT source: Github Multi-task and Self-Supervised learning Clinically useful

[85] Clinical CT scan images of 176 COVID-19 cases Random Forest / Three-fold cross-validation The Random Forest model showing promising performance for reflecting the severity of COVID-19

[86] Multiple International Datasets of CT scan images (Chinese CDC, Hospitals from China and USA, and Chainz.cn) ResNet50 High accuracy in identifying whether COVID-19 cases

[82] 4356 CT scan images (including COVID-19 and Non COVID-19 scans) collected from 6 Hospitals Present a Deep Learning model CovNET The model has the ability to high accuracy in identify COVID-19 cases from other lung diseases

[88] 618 Clinical CT scan images (including COVID-19 and Non COVID-19 scans) Deep Learning - ResNet18 Using a location attention mechanism on the model distinguishes COVID-19 cases from others with higher accuracy

[89] Clinical CT Scan Images from 99 Patients from 3 Hospitals (including COVID-19 and Non COVID-19 scans) Modified Transfer Learning - Inception Model Use a fine-tuning technique with pre-trained weights

[90] Clinical CT scan images from 133 Patients from Hospital in China Multi Stage - DL Models, LSTM The model is capable of extracting spatial and temporal information efficiently thereby improving prediction performance

[91] CT scan images of 413 COVID-19 cases and 439 of pneumonia or normal cases ResNet50 The model with transfer learning technique performs better than alternative supervised learning methods

[92] CT scan images obtained from 5 Hospitals in China Various DL learning models (Inception, ResNet50, 3D U-Net++) Able to deploy the model in 4 weeks overcoming challenges and achieving a high accuracy rate

[83] 549 CT Scan images obtained from clinics in China Deep Learning - VB-Net Introducing a Human-in the loop section to refine automation of each case - the model can segment and quantify infected regions

[93] Large-scale dataset including 10,250 CT scan images of COVID-19 and Non COVID-19 scans UNet - 2D Segmentation DL CNN Model The DL models outperforms radiologists in diagnostic performance

[94] Clinical CT scan images of 558 COVID-19 patients with pneumonia, collected from 10 hospitals COPLE-Net + Noise-Robust Dice Loss The technique presented outperforms standard Noise-Robust loss functions,COPLE-Net and the framework both perform quite highly in segmenting labels for COVID-19 pneumonia lesion

[95] Clinical CT scan images of 83 COVID-19 cases and 83 Non COVID-19 cases BigBiGAN (bi-directional generative adversarial network) The model achieves high validation accuracy in identifying COVID-19 pneumonia from CT images

[96] Large-scale dataset that include 400 COVID-19, and more Non COVID-19 scans Classification, Segmentation and Encoder-Decoder Model - Res2Net Model is highly efficient for Classification and Segmentation

[97] Multiple datasets that include 473 COVID-19 CT scans UNet Propose a method to incorporate spatial and channel attention

[98] Dataset of CT-Scans from 1,684 COVID-19 patients Inception V1 Validate the model in 3 ways including 10-fold cross-validation achieving high AUC for the validation dataset

[99] Clinical CT scan images including 146 COVID-19 cases and 149 Non COVID-19 cases DenseNet The model classifies COVID-19 over CT Images with high AUC

[34] 219 CT scan images of COVID-19 and 399 CT scan images of normal or other diseases VGG-16, GoogleNet, ResNet Support Vector Machine (SVM) was used for binary classification

[100] 746 CT scan images of COVID-19 and Non COVID-19 cases; Open-Sourced - Github Capsule Networks (CapsNets), ResNet Authors present a detail oriented capsule network, implementing data augmentation techniques to overcome lack of data

[101] Clinical CT scans of 88 confirmed patients confirmed with COVID-19 from hospitals in China DeepPneumonia (ResNet-50) Model is capable of predicting COVID-19 with high accuracy

[102] 1,129 Clinical CT scan images for COVID-19 detection UNet, Weakly Supervised DL Network (DeCovNet) Accurately predict COVID-19 infectious probability without annotating lesions for training

C. Artificial Intelligence of Things (AIoT)

In general, applications of the Internet of Things (IoT)[117]–[119] and AI can assist businesses with process automa-tion which would decrease the contacts of humans due to thelower number of people needed [120]. During the COVID-19pandemic, AI and IoT are getting more attention in thehealthcare domain where screening and detecting procedurescan be done more safely. Thermal imaging and social distancemonitoring are two main functions that are mainly consideredin the screening phase of COVID-19. In fact, the aims of thosedevices are high-temperature detection, face mask screening,and distance controlling that are discussed in the comingsections.

1) Thermal Imaging: With respect to the IoT thermalscreening applications, AI can assist in this area by implement-ing appropriate algorithms. SmartX [121], a thermal screeningdevice using infrared thermal imaging and AI face recognitionmakes screening in crowd buildings or entrances more efficient(see Figure 3). The device captures a visitor’s temperatureand also checks whether he or she is wearing a face mask.

A similar device has been developed in Taiwan for a hospitalwith the collaboration of Microsoft in order to detect facemask wearings and temperatures. Consequently, any detectionof proposed problems can easily be reported to the staff whichwill maintain an uncontaminated atmosphere [122].

Fig. 3. An industrial thermal imaging system enabled with AI [121].

TABLE IIIAI CHATBOTS/VIRTUAL ASSISTANTS COMBATING COVID-19.

Reference AI chatbot/virtual assistant Name Origin Company Function[108] Aapka Chkitsak India Academic Research Remote consultation

[109] COVID-Chatbot Tunisia & Germany Academic Research Enhancing awareness regarding COVID-19

[110] Bebot Japan Bespoke Update information & Symptoms checker

[111] Orbita USA - Reduce contacts

[112] Hyro Israel - Interacting with patients

[116] Symptoma Austria - Diagnosing by symptoms checking

[116] COVID-BOT France Clevy.io Assist with symptoms by knowledge of government and WHO

2) Social Distance Monitoring: Regarding the necessityof practicing social distancing using IoT devices, AI canimplement an automated screening using computer visionmethods [123]. An instance of such a device is RayVision[124] which ensures social distancing and face mask wearingguidelines are followed in the crowd. By using the computervision techniques, it can monitor people with a live streamon its specific dashboard which allows alerting the authoritiesin case of any rule-breaking [125]. Figure 4 represents theprocess of monitoring using cameras. Similarly, Landing AI[126] is another AI-based technology which can detect socialdistancing violations in real-time. Moreover, a peer-reviewedresearch [127] implemented an Unmanned Aerial Vehicle(UAV) or a drone with the ML application in response to theneed for maintaining social distance in crowds. Interestingly,masks will be provided by the drone for people who do notwear a mask.

Fig. 4. An industrial screening system for monitoring the social distance ofpeople and their personal protective equipment [124].

III. ML TECHNIQUES TOWARDS PREDICTING ANDTRACKING THE SPREAD OF COVID-19

As we are currently in the midst of a global pandemic, theability to predict and forecast the spread of the COVID-19could i) help the general population in taking preventativemeasures, ii) allow healthcare workers to anticipate and pre-pare for the wave of potentially infected patients, and iii) allowpolicymakers to make better decisions regards the safety of thegeneral population. More importantly, the ability to predict thespread of the virus can be used to mitigate and even prevent

the spread of COVID-19. ML models can be utilized regardingthe forecasting of the virus in providing early-signs of theCOVID-19 and projecting its spread. Also, contact tracing andsocial media data analysis have shown promising results inCOVID-19 spread mitigation [128]. The scope of this sectionis to provide a review of the research of ML tools and modelsthat can make this possible.

A. Early Signs, Preventing the Spread

Obtaining early-warning signs for an outbreak of an epi-demic could really help towards slowing and mitigating thespread of the virus. It also can encourage societies to takenecessary precautionary measures [129]. In this section, wereview the early-warning signs that were made possible usingthe ML technology. The World Health Organization (WHO)made statements about COVID-19 being a potential outbreakon the 9th of January 2020. There were AI companies likeBlueDot and Metabiota that were able to predict the outbreakeven earlier [130]. BlueDot focuses on spotting and predictingoutbreaks of infectious diseases using its proprietary methodsand tools. They use ML and NLP techniques to filter andfocus the risk of spreading a virus. Using the data fromlocal news reports of first few suspected cases of COVID-19,historical data on animal disease outbreaks, and airline ticketinformation, they were then able to use their tool to predicta definite outbreak occurring within nearing cities and otherregions of China [33]. BlueDot had warned its clients aboutthe outbreak on the 31st December 2019, over a week priorto the WHO made any statements about it [130]. Similarly,Metabiota used their ML algorithms and Big Data to predictoutbreaks and spreads of diseases, and event severity [33].They used their technology and flight data to predict thatthere will be a COVID-19 outbreak in countries like Japan,Thailand, Taiwan, and South Korea.

B. Contact Tracing

One of the major approaches for preventing the spreadof the virus is tracing the confirmed cases of COVID-19because of the potential spread of the virus through dropletsby coughs, sneeze, or talking [131]. It is recommended thatnot only the people who have a positive COVID-19 test,but also the ones who had been in close contact with the

confirmed cases to be quarantined for 14 days. The contacttracing applications are applied all over the world for thispurpose with different methods. Basically, it starts after thediagnosis process because the detected case needs to betraced. Most importantly, after the data is collected by thoseapplications, ML and AI techniques will start analyzing fordiscovering further spread of the disease [132]. Although thecontact tracing applications could be deeply helpful during thepandemic, privacy issues can bring high concerns regarding thesurveillance of individuals by some governments as a resultof huge amount of the collected data [133]. Using the digitalfootprint data provided by the applications along with MLtechnology could allow users to identify infected patients andenforce social distancing measures.

A real-time contact tracing using AI has been appliedin South Africa using the Sqreem platform [134], whichis developed in Singapore in order to track people by themetadata of the device. This information is not including thepersonal data of the user. If the user enters an infected area,he or she will be contacted by the authorities with respectto the probability of infection. Mostly, contact tracing hasbeen conducted using a major accessible device, which isa smartphone. A variety of smartphone applications enabledwith ML or AI have been adopted in order to slow the spreadof the virus by tracking and warning regarding the unsafecontacts [135]. Within the process of development, the veryfirst part would be the consideration of framework, eithercentralized or decentralized using appropriate technologiessuch as Global Positioning System, Quick Response codes,and Bluetooth [136]. ML can enable alerting automaticallyand analyzing the massive captured data, which would reducethe workforce [132]. Apple and Google announced that aBluetooth-based platform for tracking close contacts will beimplemented in the upcoming months. This technology willenable higher participation and better communication [137].An application is developed in South Korea in order to capturethe areas using location-based information, where a confirmedcase went before testing positive for COVID-19. Moreover,text messages will be sent to people who may have beenexposed due to the contamination areas [138].

Regarding the numerous implemented contact tracing ap-plications, Table IV presents some of them that have beenimplemented in various countries [139].

C. Forecasting

Forecasting epidemics centers on tracking and predictingthe spread of infectious diseases and viruses. During anepidemic, forecasting methods and models can be trained onepidemiological related data to provide an estimated numberof infected cases, patterns of spread that can give healthcareworkers guidance on how to prepare appropriately for anoutbreak [157]. Previously statistical forecasting tool such asSusceptible, Infected, Recovered models (SIR Models) havebeen used to determine the spread of a disease through thepopulation [158]. Recently, with the COVID-19 pandemic,

using ML approaches for forecasting the spread of COVID-19is getting lots of attention among the research communities.

Hu et al. [32] proposed an unsupervised ML methodfor forecasting. They used a Modified Auto-Encoder model(MAE) and trained it to predict the transmission of COVID-19cases across 31 provinces or cities in China integrating K-means algorithms to achieve a high prediction accuracy. Yanget al. [159] used epidemiological data of COVID-19 in anSEIR (Susceptible, Exposed, Infectious, Removed) model topredict the spread. They also presented another interesting ap-proach by pre-training an LSTM (Long Short-Term Memory)model which is a Recurrent Neural Network (RNN) modelon data from SARS-CoV to predict the spread of COVID-19.Their findings discovered that both models achieved similarresults in predicting the number of cases.

Similarly, another ML model with clustering techniquestrained on data from the Chinese CDC, internet searches, andnews articles create a 2-day ahead real-time forecast about thenumber of confirmed cases for 32 provinces in China [160].While in some cases there are not enough COVID-19 dataavailable for accurate forecasting using ML techniques, Liuet al. [160] overcame the lack of data by implementing dataaugmentation techniques.

Al et al. [161] introduced a novel forecasting techniquethat allows them to predict the number of confirmed casesin China over the next 10 days. The authors combined andmodified a Flower Pollination Algorithm (FPA) [162] and aSalp Swam Algorithm (SSA) [163] to improve and evaluatethe optimal parameters for an Adaptive Neuro-Fuzzy InferenceSystem (ANFIS) [164] by creating an FPASSA-ANFIS modelthat shows greater performance compared to other optimalparameters such as Root Mean Squared Relative Error (RMSE)or Mean Absolute Percentage Error (MAPE).

Alternatively, Rizk et al. [165] integrated algorithms andtechniques like Interior Search Algorithm (ISA) and ChaoticLearning (CL) into a Multi-Layer Feed-Forward Neural Net-work (MFNN) creating a forecasting model called ISACL-MFNN. Combining the two, CL and ISA, approaches im-proved the overall performance of ISA. The authors retrieveda dataset from WHO that included data from USA, Italy,and Spain between January 2020 and April 2020. Theytrained their model on this dataset and the model is thenassessed through the similarly aforementioned techniques suchas RMSE and MAPE, and more.

A new ML methodology GROOMs was proposed by Fonget al. [166] for forecasting. The authors provided an ensembleof forecasting and polynomial neural techniques that weretrained over the limited data to determine the technique thatwould yield with the lowest prediction error. Ayyoubzadehet al. [167] used Liner Regression (LR) and LSTM modelsto predict the spread of COVID-19 in Iran. The authorstrained their model on data from Google Trends [168] andWorldometer [169]. They evaluated the model with 10-foldcross-validation and use RMSE as a performance metric.Ghazaly et al. [170] utilized the limited data from WHO aboutconfirmed COVID-19 cases and deaths between January 2020

TABLE IVCONTACT TRACING APPLICATIONS COMBATING COVID-19.

Reference Application Function Origin Technology

[140] AarogyaSetuTrack close contacts of users

IndiaBluetooth

Notifying user if captured users are infected GPS location

[141] Alipay Health Code

Track close contacts of users

China

GPSDisplay the situation of user by three colors Bank transactions’ historyThe situations include healthy, in need of short or long quarantineTracking traveling information, and body temperature

[142] BeAware BahrainTrack close contacts of users

BahrainBluetooth

Quarantined and self-isolated tracking Location

[143] COVIDSafeTrack close contacts of users

Australia BluetoothNotifying user if captured users are infected

[144] CovTracerTrack close contacts of users

Cyprus GPS LocationNotifying user if captured users are infected

[145] CovidRadarTrack close contacts of users

Mexico BluetoothNotifying user if captured users are infected

[146] EhterazTrack close contacts of users

QatarBluetooth

Notifying user if captured users are infected GPS

[147] eRouska(CZ Smart Quarentine)Track close contacts of users

Czech Republic BluetoothNotifying user if captured users are infected

[148] GH Covid-19 Tracker App)Track the places an infected user had gone

Ghana GPSAllow for reporting symptoms

[149] Hamagen Track close contacts of users Israel Location based on API

[150] ImmuniTrack close contacts of users

Italy Bluetooth Low EnergyNotifying user if captured users are infected

[151] ItoMeasure the chance of infection

Germany BluetoothGuide for better safety manner

[152] Mask.irTrack close contacts of users

Iran BluetoothProvide a map of contaminated areasAllow for reporting symptoms

[153] MyTraceTrack close contacts of users

Malaysia Bluetooth Low EnergyNotifying user if captured users are infected

[154] StopCovidTrack close contacts of users

France BluetoothNotifying user if captured users are infected

[155] TraceCovidTrack close contacts of users

UAE BluetoothAccessing to the user’s information by government (privacy concern)Notifying user if captured users are infected

[156] TraceTogetherTrack close contacts of users

Singapore BluetoothAccessing to the user’s information by government (privacy concern)Notifying user if captured users are infected

and April 2020 to train a Non-Linear Aggressive Model (NAR)and predict the future cases and deaths that could occur in 9countries. However, due to not having enough historical datafor their model, the authors concluded their network is unableto continue to predict the future. Over the same time periodof those three months [171] Roy et al. implemented theirML techniques to forecast the number of cases for infected,recovered, and deceased cases country-wise and globally. Theauthors used a type of regression model called the ProphetPrediction Model. Developed by Facebook, this model has acapacity to create a precise time-series forecast that is simpleto create and could result in accurate prediction.

Bandyopadhyay et al. [172] used a RNN and GRU (GatedRecurrent Unit) model to predict the number of confirmedcases and deaths. The model was trained on data sourcedfrom Kaggle on confirmed cases between January 2020 andMarch 2020. The results presented in their findings indicatethat the RNN is capable of predicting the cases and assessingthe severity of COVID-19. Utilizing the ML techniques, the

Global Virus Tracker [173] built up a system embedded thelocation and symptom risk evaluation for the users. They alsodeveloped the chatbot to provide the conversational interfacefor the same ML-based evaluation. In the proposed system,the county or city level risk, population density and updatedvirus spreading were incorporated at the moment of the riskevaluation.

D. Social Media Analysis

Social media has become a platform where people sharepictures, reviews, posts, and exchange stories. A popular socialmedia platform where people may obtain and access news isTwitter. Its users can validate live alerts and obtain informationdirectly through the smartphone application. This is possibleas major news outlets, government bodies, community centers,etc., all have accounts that they use to share updates onTwitter. Users can also use the platform to share their personalexperiences via tweets. It can essentially be considered aform of microblogging for users who just want to share their

insight over a certain topic. Tweets can be a form of datathat can analyze feedback and obtain public sentiment overcertain topics. Over the course of the COVID-19 pandemic,people have engaged on twitter to share their experience withregards to testing or lack of testing, social-distancing, andother challenges that people are facing due to the pandemic[174]. In the remainder of the subsection, we review theresearch dedicated on how ML technology can be used toanalyze social media for COVID-19 related updates.

The COVID-19 was declared a global pandemic in March2020. However, people had already been posting and dis-cussing it over social platforms. Between January 27th andMarch 26th, 2020 there were 5,621,048 tweets with keywords“corona virus” and “coronavirus” according to Karisani et al.[175] when they constructed a dataset for their ML models.The dataset consists of both positive and negative tweetsshared by users. Using this dataset, the authors were able toimplement several ML methods like Logistic Regression (LR),Naive Bayes Classification (NB) to automate the detection ofCOVID-19 positive results shared over twitter by users. BothLR and NB supervised ML models were also used by Samuelet al. [176] to obtain public sentiment and feedback aboutCOVID-19 through users tweets shared over the platform. Asthere is an abundance of news and tweets shared over twitter,a sentiment analysis was done by [177] addressing the needfor filtering it out as there is a potential of misinformationbeing spread across social platforms. The authors implementedan unsupervised ML topic modelling technique known asLatent Dirichlet Allocation (LDA). The use of LDA is doneto spot the semantic relationship between words in a tweetand provides a sentiment analysis on whether the tweetsare positive and showing signs of comfort or whether theyare negative showing discomfort and panic. In their findings,negative tweets are higher as people are prevalent in anger andsadness towards quarantine and death.

Jahanbin et al. [178] gathered data from Twitter by search-ing for COVID-19 related hashtags. The dataset of tweets ispre-processed and filtered first to reduce the noise by removingirrelevant data. This would allow training a better model forclassification. The authors used an evolutionary algorithmEclass1-MIMO that predicts the morbidity rate in regions.Mackey et al. [179] introduced an unsupervised ML approachthat analyzes tweets by users who may be infected by thevirus, recovering from it or the experiences they had related totesting for it. The authors used a Biterm Topic Model (BTM)combined with clustering techniques to determine statisticaland geographical characteristics depending on content analy-sis. Obtaining feedback over Twitter and social media outletswill give a live reflection of how the general public is reactingto the pandemic and will allow policymakers in making betterdecisions.

IV. ML TECHNIQUES TOWARDS MEDICAL ASSISTANCE

As the virus spreads across the world infecting more ofthe population and with the death toll rising rapidly, effortsare made to develop an effective vaccine or discover a drug

for COVID-19. The immunologists around the world surgedtowards studying the symptoms, and the immune responsesthat infected patients show in relation to combating the virus[184]. In this section, we review the efforts of the scientificcommunity in using ML techniques regarding understandingthe virus [52], how to attack it, and perhaps even be able tofind a cure for COVID-19.

A. Understanding the Virus

Analyzing the genomics and proteomics characteristics ofa viral disease is an important step to combat the disease.Scientists have been studying the virology of COVID-19which can give the physical and chemical properties, cellentry and receptor interaction, and the overall ecology andthe genomics of the virus [185]. A genome is the completegenetic information that provides the architecture of a virusand knowing the genome for COVID-19 can help in betterunderstanding the transmissibility and infectiousness of thevirus [186]. The study of proteomics is knowing the proteinsof an organism. Identifying the proteins of COVID-19 wouldallow a better understanding of the overall protein structureand discover how the proteins would interact with the drugs[187]. Over recent years, there have been remarkable advance-ments by scientists in interdisciplinary fields of bioinformaticsand computational medicine and ML techniques have shownmeaningful interpretation towards determining genomics andprotein structures of various diseases [188]. In this section, wefocus on COVID-19 and discuss the ML techniques that havebeen implemented regarding the research of interpreting thegenomics and proteomics of that.

COVID-19 is an RNA (ribonucleic acid) type of virus fromthe coronavirus (CoV) family. It is a single-stranded RNA witha large viral genome. These large genomes can have two orthree viral proteases. For COVID-19 it has 2 proteases, whichwe will refer to as 3CLpro. COVID-19 belongs to the samefamily as the aforementioned respiratory diseases SARS andMERS [189]. Viruses that belong to this type of family caninfect a range of animal species such as camels, cats, cattle,bats, as well as humans [190]. They are easily transmittableand can infect a host in one species and transmit it ontoanother species [190]. Multiple findings suggest that the originof COVID-19 in infecting humans is transmitted from batswith an 89% similarity structure identity to a coronavirus thatinfects bats (SARS-like-CoVZXC21) [191].

Identifying and determining the biological structure at amolecular level of a viral disease is important to developan effective therapy towards finding a cure for the disease.However, this process involves a lot of experiments that cantake months. Computational ML techniques and models canspeed up this process and predict the structure of proteinsaccurately [192].

The family of coronavirus has multiple classes, and thevirus can belong to either the alpha or the beta class ofvirus. SARS-CoV and MERS both were determined to bebeta coronaviruses [193]. To determine and classify whatspecific type of virus COVID-19 could be, Randhawa, et al.

TABLE VPREDICTIVE ANALYSIS TOOLS AND METHODS COMBATING COVID-19.

Reference Section Model and Technology Remarks[33]

Early Tracking, PreventionPredictive Analysis tool Using flight details data and recent outbreaks to predict the spread in nearby countries

[32] ANN - K-Means Algorithm Using a MAE to successfully predict 2-day spread[170]

Forecasting

Non-Auto Regressive Neural Network Prediction Error, due to scarcity of error at the time of Analysis[160] Augmented ARGONet Clustering of Chinese Provinces, and getting a 2-day forecast[166] Polynomial Neural Network (PNN) - GROOMS Addressing data augmentation and importance of early forecast[172] RNN Researching predicting using GRU + LSTM combined models[180] K-Means Clustering Algorithm Possible to predict the spread of cases[161] FPASSA-ANFIS (ANN) Predict a 10-day forecast of the number of cases in China[165] ISACL-MFNN Predict a 10-day forecast of the number of cases in multiple countries[167] LSTM and LR models Predict and forecast of the number of cases of COVID-19 in Iran[171] Regression Model, Prophet Prediction Time-Series Forecasting[181]

Social Media Analysis

AI Algorithms Phone based survey to determine whether a person is high-risk, low-risk or contracting the virus[178] Eclass1-MIMO Classifying a twitter dataset to determine morbidity in regions[176] Natural Processing Language Getting public sentiment by classifying tweets[177] Latent Dirichlet Allocation (LDA) Algorithms used to spot semantic relationship between words[182] Sentiment Analysis Building a visual cluster to highlighting public opinion over pandemic[179] Unsupervised ML (biterm topic model) Attain content analysis by assessing user tweets[175] Naive Bayes, Logistic Regression, and more. Automate detection of positive COVID-19 report results through tweets[183] Shallow Neural Networks Training multiple word2vec models to put context to words

[194] used a supervised ML technique with digital signalprocessing techniques (MLDSP) to evaluate the taxonomy ofthe virus. MLDSP techniques have previously been used toachieve high accuracy in the classification of other viruses anddiseases such as HIV-1 genomes and Influenza [195]. Usingthis model, the authors could evaluate that like its predecessor,COVID-19 also belongs to the beta coronavirus. To predictthe protein structure of COVID-19, Heo and Feig [196] useda ML-based method called TrRosetta. This method can beused to predict the inter-residue distance and create structuremodels for the protein. To do this with higher predictionefficiency, the authors applied molecular dynamics simulation-based refinement.

Magar et al. [197] highlighted the importance of biologicalstructure and protein sequences in combating the virus. Theauthors developed an ML model to predict how synthetic anti-bodies inhibit the spread of COVID-19. The model can predictthe response of these antibodies by understanding the bindingbetween the antibodies and the viral mutations. Training themodel on a dataset that includes virus anti-body sequences andthe clinical patient neutralization response allowed the authorsto predict antibody responses. The authors used ML techniquessuch as XGBoost, RF, Multilayer Perceptron (MLP), SVM,and LR for their model. The XGBoost model achieves thehighest accuracy in prediction over an 80%-20% split of trainand test data.

B. Drug and Vaccine Development

As COVID-19 cases continue to rise numbers of both theinfected cases and the death toll, it has become an urgentneed to discover a drug that could mitigate these numbersfrom increasing any further. ML techniques can be used toanalyze how drugs react to the viral proteins of COVID-19.We have already seen ML methods and techniques like SVMand Bayesian Classifiers being used for drug discovery andrepurposing [198]. In this section we review the ML studiesand research that had been done about discovering the new

drugs or repurposing the currently approved FDA ones. Wealso review the ML research that has been done regarding thevaccine development.

1) Drug Development and Repurposing: An exploratoryapproach of determining whether commercially available anti-viral drugs can treat or help towards reducing the severityof COVID-19 infected patients was presented by Beck etal. [199]. They used a pre-trained ML learning model calledMolecule Transformer-Drug Target Interaction (MT-DTI), aninteraction prediction model, to predict the binding affinitybetween COVID-19 infected proteins and compounds. Theobjective of their study was to identify potential FDA approveddrugs that may restrain the proteins of COVID-19. MT-DTIis capable of predicting the chemical sequences and aminoacid sequences of a target protein without the whole structureinformation. This is helpful to use as there was limited knowl-edge on the overall structure of viral proteins of COVID-19initially. Regarding their advantage, the authors used theMT-DTI model to predict binding affinities of 3,410 FDA-approved drugs. Similarly, Heiser et al. [35] used proprietaryDL techniques for the purpose of drug discovery. They usedtheir model to evaluate how FDA and European MedicinesAgency (EMA) approved drugs and compounds would affecthuman cells, analyzing over 1,660 drugs.

In some cases, various drugs from antiviral to antimalarialcould be used to combat COVID-19 [200]. These drugs usedfor combating severe illnesses are referred to as “parents”by Moskal et al. [200] in their study. The authors used MLtechniques like CNN, LSTM, and MLP analyzing the molec-ular similarity between these “parents” drugs and second-generation drugs that could potentially be also used to fightagainst the virus. The authors introduced the second generationdrugs as “progeny”. This study is important in predicting otherdrugs that may be helpful in this pandemic. It can result inhaving a larger catalog of drugs, which provides alternativesolutions if the “parent” drug fails to respond. To convert the

molecular structure into a high-dimensional vector space, theMol2Vec [201] method was used.

Kadioglu et al. [202] identified three viral proteins as targetsfor their ML approach. They targeted the Spike protein, thenucleocapsid protein, and the 2’-o-ribose-methyltransferaseprotein. The spike protein acts as a cellular receptor for thehost of the virus. The nucleocapsid protein plays a vital rolein coronavirus transcription and the overall forming of thegenomics of the RNA virus. The 2’-o-ribose-methyltransferaseprotein is an essential protein for coronavirus synthesis andprocessing. The authors used ML algorithms against the threeproteins in predicting how FDA-approved drugs and naturalcompounds react to the three proteins with such key charac-teristics.

In [203] the authors used a DNN to predict and generatea new small design for molecules that would be capableof inhibiting COVID-19 3CLpro. Targeting 3CLpro can bean essential part with respect to the drugs development forCOVID-19. Alternatively, Zhavoronkov et al. [204] utilized28 various types of ML methods such as Autoencoders,Generative Adversial Networks, Genetic Algorithms to predictand generate the molecular structures. Using deep-Q learningnetworks, Tang et al. [205] were able to generate 3CLprocompounds of COVID-19. Being able to successfully predictthese protein targets can provide advancements in developinga potential drug for the virus. Hu et al. [206] created an MLmodel that predicts the binding between the potential drugsand COVID-19 proteins.

2) Vaccine Development: Once a virus starts to spread andturn into a global pandemic, there is a very little chanceof stopping it without a vaccine [207]. That stands true forCOVID-19 as well. Historically, vaccination has been thesolution to control or slow the spread of a viral infection[208] [209]. It is critical to have a vaccine developed toprovide immunity against COVID-19 and stop this pandemic.So far, the research for vaccine development of COVID-19is dedicated with three different types of vaccines [210]. TheWhole Virus Vaccine represents a classical strategy for thedevelopment of vaccinations of viral disease. Subunit Vaccinerelies on extracting the immune response against the S-spikeprotein for COVID-19 [210]. This will refrain it from dockingit with the hosts receptor protein [210]. The Nucleic AcidVaccine produces a protective immunological response to fightagainst the virus by [211].

At present, there are over 22 vaccines that are at the clinicalstage of trials in combating COVID-19 [212]. The process ofdeveloping a vaccine would need to first go through the designstage and then towards the testing and experimental stage onanimals and eventually in humans. In this section, we reviewthe implicit research that is done over vaccine developmentand how ML techniques have been employed.

Ge et al. [214] used ML techniques to evaluate how smallvirus strings (called peptides) bind to the human receptorcells. The authors created two ML programs OptiVax andEvalVax. Optivax combines different pre-existing programs

that evaluate combinations of peptides and receptor cells. Aprogram that OptiVax adopts is NetMHCPan [222], which usesa feed-forward neural network. Thereby, peptide informationand receptor cell are input data that generate an affinity scoreto predict the binding as output. The prediction score isconstantly improved as it uses the back-propagation method.EvalVax utilizes the genetic ancestry of three different cat-egories of the population, i.e. white, black, and Asian. Thisdata is used as an objective function to discover peptide andreceptor pairs for OptiVax.

Similarly, Herst et al. [220] in their research about findinga vaccine employed a similar technique that was previouslyused for combating the Ebola virus. They used ML techniqueslike Artificial Neural Network (ANN), SVM, netMHC, andnetMHCpan to predict potential vaccine candidates.

Ward et al. [215] mapped out the protein sequences ofCOVID-19. This data was used by the authors for prediction,specificity, and epitope analysis. Epitope is the molecule thatadds antibodies attached and is recognized by the immunesystem. The authors used the ML-based program NetMHC-Pan to locate the epitope sequences. With this, the authorswere able to create an online tool that would aid in epitopeprediction, coronavirus homology analysis, variation analysis,and proteome analysis. Another study about epitope predictionwas presented by Qiao et al. [219]. They employed DLtechniques that are able to predict the best epitope for peptide-based COVID-19 vaccinations [219]. To do this, the authorsproposed a DL model that utilizes LSTM and RNN methods tocapture sequence patterns of peptides and eventually be able topredict the new mutant antigens for patients. As an alternativeapproach to predicting the epitope and protein sequence, anML based tool called Ellipro was utilized by Rahman et al.[216]. The tool is capable of predicting and presenting avisual view of the protein sequence of the epitope within thestructure. Similarly Sarkar et al. [217] used the SVM methodto predict the toxic level of some epitopes.

The study towards epitope prediction continues in the re-search done by Prachar et al. [218] who employed varioustechniques such as ANN and Position-Specific Weight Ma-trices (PSSM) algorithms to predict and verify COVID-19epitopes. Ong et al. [213] introduced a vaccine designingapproach referred to as Reverse Vaccinology (RV). The aimof RV is to identify a potential vaccine through bioinformaticsanalysis of the pathogen genomes. They used Supervised MLclassifications such as LR, SVM, K-nearest Neighbor, RF,and Extreme Gradient Boosting (XGB) to train on annotatedproteins dataset with the objective of getting high predictionaccuracy for the protein candidates, the authors used an MLbased tool referred as Vaxign-ML.

V. FUTURE EXPECTATIONS

Although we review many of the ML approaches regardingthe impact of COVID-19 in this paper, there is still anessential need for developing solutions using ML to addressthe pandemic’s complications and challenges. Since there wasno adopted method for fighting against COVID-19 when it

TABLE VIVACCINE AND DRUG DEVELOPMENT OF COVID-19 USING ML ALGORITHMS.

Reference Sections Model and Technology Remarks[213]

Vaccine Development

ML Algorithms Logistic Regression, Support Vector Machine, etc Using a tool called Vaxign to implement Reverse Vaccinology[214] OptiVax, EvalVax, netMHCpan, etc. Predicting binding between virus proteins and Host cell Receptors[215] NetMHCPan Create an online tool for visualisation and extraction of COVID-19 meta-analysis[216] Ellipro, multiple ML methods Predict the epitope structure[217] SVM Review epitope-based design for a COVID-19 vaccine[218] ANN Predicting COVID-19 epitopes[219] DeepNovo, LSTM, RNN Predictive analysis of protein sequences to discover antibodies in patients[220] netMHCpan, netMHC Use ML techniques to predict peptide sequences[35]

Drug Re-purposing

DL Models - Neural Networks Analysing how approved FDA Drugs would work against COVID-19[199] DL - Drug Target Interactions Re-purposing current drugs to find out whether there is an affinity between drug and proteins[202] Neural Networks and Naive Bayes Predicting drug interaction between proteins and compounds[200] CNN, LSTM and MLP models ML techniques to predict similarities between available drugs to combat COVID-19[206] Fine-Tuning AtomNet based Model Predicting binding between COVID-19 proteins and drug compounds[204] Various ML methods Use RL strategies to generate new 3CLpro structure[203] Deep Neural Network Creating small molecule interaction and targeting 3CLpro[221] Deep Learning Models Provide large scale virtual screening to identify protein interacting pairs

was started, previous ML models regarding infectious diseases(epidemiological models) can be helpful for the early stageof COVID-19. As we discussed, detecting and screeningCOVID-19 using AI and ML techniques can play a key rolein combating this pandemic. The combination of technologieslike IoT devices with these techniques needs to be expandedfor crowd areas including airports, subways or bus stations,and so on. This development would enhance the identificationof suspicious cases within lesser both time and contamination.It is important to implement an efficient method of achiev-ing high accuracy of detecting COVID-19 through medicalimaging and integrating ML techniques. It is essential toovercome the challenges that are presented by this approach.Challenges such as lack of data and the privacy issue withindata collection, false information by media, the limited numberof expertise between AI and medical science. AI technologiescan also assist to implement the following expectations inthe future: i) empowering the medical imaging devices usingthe noncontact automatic image capturing to prevent furtherinfection from the patient to the radiologist or even anotherpatient, and ii) automatically monitoring the patients usingintelligent video analysis. As the countries learn more aboutCOVID-19, it is essential to have updated datasets. This canlead to better forecasting by implementing the proper MLmodels using those datasets. In addition to the forecastingconcepts, investigating the effects of different social mediaand their pathways for detecting early sign of possible futurepandemic.

VI. CONCLUSION

Machine Learning (ML) models and techniques have vastlybeen used in plenty of industries over the past decade.Within the healthcare industry, ML has been usually usedfor screening and diagnosing. In epidemiology area, ML isbasically utilized for forecasting and understanding epidemicsand diseases. In this paper, we presented a comprehensivesurvey of how ML applications have been used to fight againstCOVID-19. We presented the efforts that are taken by the MLresearch communities to combat this virus across three mainphases “Screening”, “Tracking and Forecasting” and “Medical

Assistance”. ML applications for each of these phases areprimarily focused as such; “Screening” intended for diag-nosing the virus through medical imaging data (COVID-19related X-Rays and CT-Scans), “Tracking and Forecasting”towards forecasting and predicting the numbers of cases andcontact tracing, and lastly, “Medical Assistance” with the aimof understanding the protein sequences and structure of thevirus and whether a cure could be found in combating it via adrug or vaccine. One of the main challenges that researchersface when diagnosing using ML techniques was the lack ofrelevant data that are made accessible to the public. Lackof data meant researchers had to use techniques like dataaugmentation, transfer learning, and fine-tuning models toimprove prediction accuracy. Though these methods workedwell in some cases, more data would make these models morerobust. Similarly, forecasting models trained on more data forpredicting the spread and number of cases could be moreaccurate. Regarding developing a vaccine or re-purposing,it is important to have a good understanding of virology,bioinformatics. Additionally, ML is especially important forresearchers from different fields to collaborate and integratetheir knowledge in order to discover a cure.

REFERENCES

[1] S. Latif, M. Usman, S. Manzoor, W. Iqbal, J. Qadir, G. Tyson, I. Castro,A. Razi, M. N. K. Boulos, A. Weller et al., “Leveraging data scienceto combat COVID-19: A comprehensive review,” 2020.

[2] L. Morawska and J. Cao, “Airborne transmission of SARS-CoV-2: Theworld should face the reality,” Environment International, p. 105730,2020.

[3] CDC. Symptoms of coronavirus. [Online]. Avail-able: https://www.cdc.gov/coronavirus/2019-ncov/symptoms-testing/symptoms.html

[4] H. Nishiura, T. Kobayashi, T. Miyama, A. Suzuki, S.-m. Jung,K. Hayashi, R. Kinoshita, Y. Yang, B. Yuan, A. R. Akhmetzhanov et al.,“Estimation of the asymptomatic ratio of novel coronavirus infections(COVID-19),” International Journal of Infectious Diseases, vol. 94, p.154, 2020.

[5] J. A. Al-Tawfiq, “Asymptomatic coronavirus infection: MERS-CoVand SARS-CoV-2 (COVID-19),” Travel medicine and infectious dis-ease, 2020.

[6] G. News. Coronavirus (COVID-19). [Online]. Available: https://news.google.com/covid19/map?hl=en-US&gl=US&ceid=US:en

https://www.cdc.gov/coronavirus/2019-ncov/symptoms-testing/symptoms.htmlhttps://www.cdc.gov/coronavirus/2019-ncov/symptoms-testing/symptoms.htmlhttps://news.google.com/covid19/map?hl=en-US&gl=US&ceid=US:enhttps://news.google.com/covid19/map?hl=en-US&gl=US&ceid=US:en

[7] M. Nasajpour, S. Pouriyeh, R. M. Parizi, M. Dorodchi, M. Valero, andH. R. Arabnia, “Internet of Things for current COVID-19 and futurepandemics: An exploratory study,” arXiv preprint arXiv:2007.11147,2020.

[8] R. Vaishya, M. Javaid, I. H. Khan, and A. Haleem, “Artificial In-telligence (AI) applications for COVID-19 pandemic,” Diabetes &Metabolic Syndrome: Clinical Research & Reviews, 2020.

[9] A. Alimadadi, S. Aryal, I. Manandhar, P. B. Munroe, B. Joe, andX. Cheng, “Artificial Intelligence and Machine Learning to fightCOVID-19,” 2020.

[10] N. L. Bragazzi, H. Dai, G. Damiani, M. Behzadifar, M. Martini, andJ. Wu, “How Big Data and Artificial Intelligence can help better man-age the COVID-19 pandemic,” International Journal of EnvironmentalResearch and Public Health, vol. 17, no. 9, p. 3176, 2020.

[11] M. C. Chang and D. Park, “How can Blockchain help people inthe event of pandemics such as the COVID-19?” Journal of MedicalSystems, vol. 44, pp. 1–2, 2020.

[12] M. E. Darnell, K. Subbarao, S. M. Feinstone, and D. R. Taylor,“Inactivation of the coronavirus that induces severe acute respiratorysyndrome, SARS-CoV,” Journal of Virological Methods, vol. 121,no. 1, pp. 85–91, 2004.

[13] M. Willman, D. Kobasa, and J. Kindrachuk, “A comparative analysisof factors influencing two outbreaks of middle eastern respiratorysyndrome (MERS) in Saudi Arabia and South Korea,” Viruses, vol. 11,no. 12, p. 1119, 2019.

[14] WHO. SARS - WHO. [Online]. Available: https://www.who.int/ith/diseases/sars/en/

[15] L. Jia, Q. Zhao, and L. Tong, “Retail pricing for stochastic demandwith unknown parameters: An online Machine Learning approach,” in2013 51st Annual Allerton Conference on Communication, Control,and Computing (Allerton). IEEE, 2013, pp. 1353–1358.

[16] K. Chitra and B. Subashini, “Data Mining techniques and its applica-tions in banking sector,” International Journal of Emerging Technologyand Advanced Engineering, vol. 3, no. 8, pp. 219–226, 2013.

[17] M. Chen, Y. Hao, K. Hwang, L. Wang, and L. Wang, “Disease predic-tion by Machine Learning over big data from healthcare communities,”IEEE Access, vol. 5, pp. 8869–8879, 2017.

[18] J. Clemente, F. Li, M. Valero, and W. Song, “Smart seismic sensingfor indoor fall detection, location, and notification,” IEEE journal ofbiomedical and health informatics, vol. 24, no. 2, pp. 524–532, 2020.

[19] S. Ekins, “The next era: Deep Learning in pharmaceutical research,”Pharmaceutical Research, vol. 33, no. 11, pp. 2594–2603, 2016.

[20] I. El Naqa and M. J. Murphy, “What is Machine Learning?” pp. 3–11,2015.

[21] E. Alpaydin, Introduction to Machine Learning. MIT press, 2020.[22] S. Wong, H. Al-Hasani, Z. Alam, and A. Alam, “Artificial Intelligence

in radiology: how will we be affected?” European Radiology, vol. 29,no. 1, pp. 141–143, 2019.

[23] A. D. T. Force*., M. Singer, C. Deutschman, C. Seymour, M. Wildman,C. Sanderson, J. Groves, M. Wildman, C. Sanderson, J. Groves et al.,“Artificial Intelligence in healthcare: past, present and future,” Journalof the Intensive Care Society, vol. 20, no. 3, pp. 268–273, 2019.

[24] K. Kourou, T. P. Exarchos, K. P. Exarchos, M. V. Karamouzis, andD. I. Fotiadis, “Machine Learning applications in cancer prognosisand prediction,” Computational and Structural Biotechnology Journal,vol. 13, pp. 8–17, 2015.

[25] Q. Zou, K. Qu, Y. Luo, D. Yin, Y. Ju, and H. Tang, “Predicting diabetesmellitus with Machine Learning techniques,” Frontiers in Genetics,vol. 9, p. 515, 2018.

[26] C.-C. Wu, W.-C. Yeh, W.-D. Hsu, M. M. Islam, P. A. A. Nguyen, T. N.Poly, Y.-C. Wang, H.-C. Yang, and Y.-C. J. Li, “Prediction of fatty liverdisease using Machine Learning algorithms,” Computer Methods andPrograms in Biomedicine, vol. 170, pp. 23–29, 2019.

[27] H. Asri, H. Mousannif, H. Al Moatassime, and T. Noel, “UsingMachine Learning algorithms for breast cancer risk prediction anddiagnosis,” Procedia Computer Science, vol. 83, pp. 1064–1069, 2016.

[28] M. A. Ahmad, C. Eckert, and A. Teredesai, “Interpretable MachineLearning in healthcare,” in Proceedings of the 2018 ACM internationalconference on bioinformatics, computational biology, and health infor-matics, 2018, pp. 559–560.

[29] R. Sandhu, S. K. Sood, and G. Kaur, “An intelligent system forpredicting and preventing MERS-CoV infection outbreak,” The Journalof Supercomputing, vol. 72, no. 8, pp. 3033–3056, 2016.

[30] S. Choi, J. Lee, M.-G. Kang, H. Min, Y.-S. Chang, and S. Yoon,“Large-scale Machine Learning of media outlets for understandingpublic reactions to nation-wide viral infection outbreaks,” Methods,vol. 129, pp. 50–59, 2017.

[31] S. Debnath, D. P. Barnaby, K. Coppa, A. Makhnevich, E. J. Kim,S. Chatterjee, V. Tóth, T. J. Levy, M. d Paradis, S. L. Cohen et al., “Ma-chine Learning to assist clinical decision-making during the COVID-19pandemic,” Bioelectronic Medicine, vol. 6, no. 1, pp. 1–8, 2020.

[32] Z. Hu, Q. Ge, L. Jin, and M. Xiong, “Artificial Intelligence forecastingof COVID-19 in china,” arXiv preprint arXiv:2002.07112, 2020.

[33] Z. Allam, G. Dey, and D. S. Jones, “Artificial intelligence (AI) providedearly detection of the coronavirus (COVID-19) in China and willinfluence future urban health policy internationally,” AI, vol. 1, no. 2,pp. 156–165, 2020.

[34] U. Ozkaya, S. Ozturk, and M. Barstugan, “Coronavirus (COVID-19)classification using deep features fusion and ranking technique,” arXivpreprint arXiv:2004.03698, 2020.

[35] K. Heiser, P. F. McLean, C. T. Davis, B. Fogelson, H. B. Gordon,P. Jacobson, B. L. Hurst, B. J. Miller, R. W. Alfa, B. A. Earnshaw et al.,“Identification of potential treatments for COVID-19 through artificialintelligence-enabled phenomic analysis of human cells infected withSARS-CoV-2,” bioRxiv, 2020.

[36] V. M. Corman, O. Landt, M. Kaiser, R. Molenkamp, A. Meijer, D. K.Chu, T. Bleicker, S. Brünink, J. Schneider, M. L. Schmidt et al.,“Detection of 2019 novel coronavirus (2019-nCoV) by real-time rt-pcr,” Eurosurveillance, vol. 25, no. 3, p. 2000045, 2020.

[37] X. Xie, Z. Zhong, W. Zhao, C. Zheng, F. Wang, and J. Liu, “Chestct for typical 2019-nCoV pneumonia: relationship to negative rt-pcrtesting,” Radiology, p. 200343, 2020.

[38] C. Gollier and O. Gossner, “Group testing against COVID-19,” CovidEconomics, vol. 2, 2020.

[39] Q.-V. Pham, D. C. Nguyen, W.-J. Hwang, P. N. Pathirana et al.,“Artificial Intelligence (AI) and Big Data for coronavirus (COVID-19)pandemic: A survey on the state-of-the-arts,” 2020.

[40] S. Lalmuanawma, J. Hussain, and L. Chhakchhuak, “Applications ofMachine Learning and Artificial Intelligence for COVID-19 (SARS-CoV-2) pandemic: A review,” Chaos, Solitons & Fractals, p. 110059,2020.

[41] W. Yang and F. Yan, “Patients with rt-pcr-confirmed COVID-19 andnormal chest ct,” Radiology, vol. 295, no. 2, pp. E3–E3, 2020.

[42] S. H. Kassani, P. H. Kassasni, M. J. Wesolowski, K. A. Schneider, andR. Deters, “Automatic detection of coronavirus disease (COVID-19)in x-ray and ct images: A Machine Learning-based approach,” arXivpreprint arXiv:2004.10641, 2020.

[43] A. Giovagnoni, “Facing the COVID-19 emergency: we can and wedo.” La Radiologia Medica, pp. 1–2, 2020.

[44] D. Cozzi, M. Albanesi, E. Cavigli, C. Moroni, A. Bindi, S. Luvarà,S. Lucarini, S. Busoni, L. N. Mazzoni, and V. Miele, “Chest x-rayin new coronavirus disease 2019 (COVID-19) infection: findings andcorrelation with clinical outcome,” La Radiologia Medica, p. 1, 2020.

[45] A. E. Hassanien, L. N. Mahdy, K. A. Ezzat, H. H. Elmousalami,and H. A. Ella, “Automatic x-ray COVID-19 lung image classificationsystem based on multi-level thresholding and support vector machine,”medRxiv, 2020.

[46] P. K. Sethy and S. K. Behera, “Detection of coronavirus disease(COVID-19) based on deep features,” Preprints, vol. 2020030300, p.2020, 2020.

[47] I. Castiglioni, D. Ippolito, M. Interlenghi, C. B. Monti, C. Salvatore,S. Schiaffino, A. Polidori, D. Gandola, C. Messa, and F. Sardanelli,“Artificial Intelligence applied on chest x-ray can aid in the diagnosis ofCOVID-19 infection: a first experience from lombardy, Italy,” medRxiv,2020.

[48] J. Zhang, Y. Xie, Y. Li, C. Shen, and Y. Xia, “COVID-19 screeningon chest x-ray images using deep learning based anomaly detection,”arXiv preprint arXiv:2003.12338, 2020.

[49] L. Wang and A. Wong, “COVID-Net: A tailored deep convolutionalneural network design for detection of COVID-19 cases from chestx-ray images,” arXiv preprint arXiv:2003.09871, 2020.

[50] I. D. Apostolopoulos and T. A. Mpesiana, “COVID-19: automatic de-tection from x-ray images utilizing transfer learning with ConvolutionalNeural Networks,” Physical and Engineering Sciences in Medicine,p. 1, 2020.

https://www.who.int/ith/diseases/sars/en/https://www.who.int/ith/diseases/sars/en/

[51] B. Ghoshal and A. Tucker, “Estimating uncertainty and interpretabilityin deep learning for coronavirus (COVID-19) detection,” arXiv preprintarXiv:2003.10769, 2020.

[52] J. Chen, K. Li, Z. Zhang, K. Li, and P. S. Yu, “A survey on applicationsof Artificial Intelligence in fighting against COVID-19,” arXiv preprintarXiv:2007.02202, 2020.

[53] M. Farooq and A. Hafeez, “COVID-resnet: A Deep Learning frame-work for screening of COVID19 from radiographs,” arXiv preprintarXiv:2003.14395, 2020.

[54] E. E.-D. Hemdan, M. A. Shouman, and M. E. Karar, “COVIDx-net:A framework of Deep Learning classifiers to diagnose COVID-19 inx-ray images,” arXiv preprint arXiv:2003.11055, 2020.

[55] H. Hirano, K. Koga, and K. Takemoto, “Vulnerability of deep neuralnetworks for detecting COVID-19 cases from chest x-ray images touniversal adversarial attacks,” arXiv preprint arXiv:2005.11061, 2020.

[56] P. Afshar, S. Heidarian, F. Naderkhani, A. Oikonomou, K. N. Platan-iotis, and A. Mohammadi, “COVID-caps: A capsule network-basedframework for identification of COVID-19 cases from x-ray images,”arXiv preprint arXiv:2004.02696, 2020.

[57] L. Sarker, M. M. Islam, T. Hannan, and Z. Ahmed, “COVID-DenseNet:A Deep Learning architecture to detect COVID-19 from chest radiologyimages,” 2020.

[58] F. Ucar and D. Korkmaz, “COVIDiagnosis-Net: Deep Bayes-SqueezeNet based diagnostic of the coronavirus disease 2019 (COVID-19) from x-ray images,” Medical Hypotheses, p. 109761, 2020.

[59] S. Khobahi, C. Agarwal, and M. Soltanalian, “CoroNet: A deepnetwork architecture for semi-supervised task-based identification ofCOVID-19 from chest x-ray images,” medRxiv, 2020.

[60] E. Luz, P. L. Silva, R. Silva, and G. Moreira, “Towards an efficientDeep Learning model for COVID-19 patterns detection in x-ray im-ages,” arXiv preprint arXiv:2004.05717, 2020.

[61] A. Narin, C. Kaya, and Z. Pamuk, “Automatic detection of coronavirusdisease (COVID-19) using x-ray images and Deep ConvolutionalNeural Networks,” arXiv preprint arXiv:2003.10849, 2020.

[62] H. S. Maghdid, A. T. Asaad, K. Z. Ghafoor, A. S. Sadiq, and M. K.Khan, “Diagnosing COVID-19 pneumonia from x-ray and ct imagesusing Deep Learning and transfer learning algorithms,” arXiv preprintarXiv:2004.00038, 2020.

[63] M. Toğaçar, B. Ergen, and Z. Cömert, “COVID-19 detection usingDeep Learning models to exploit social mimic optimization and struc-tured chest x-ray images using fuzzy color and stacking approaches,”Computers in Biology and Medicine, p. 103805, 2020.

[64] T. Ozturk, M. Talo, E. A. Yildirim, U. B. Baloglu, O. Yildirim,and U. R. Acharya, “Automated detection of COVID-19 cases usingdeep neural networks with x-ray images,” Computers in Biology andMedicine, p. 103792, 2020.

[65] S. Rajaraman and S. Antani, “Training Deep Learning algorithms withweakly labeled pneumonia chest x-ray data for COVID-19 detection,”medRxiv, 2020.

[66] D. N. Vinod and S. Prabaharan, “Data science and the role of ArtificialIntelligence in achieving the fast diagnosis of COVID-19,” Chaos,Solitons & Fractals, p. 110182, 2020.

[67] M. Jamil, I. Hussain et al., “Automatic detection of COVID-19 infec-tion from chest x-ray using Deep Learning,” medRxiv, 2020.

[68] S. Toraman, T. B. Alakuş, and İ. Türkoğlu, “Convolutional capsnet:A novel artificial neural network approach to detect COVID-19 dis-ease from x-ray images using capsule networks,” Chaos, Solitons &Fractals, p. 110122, 2020.

[69] M. E. Chowdhury, T. Rahman, A. Khandakar, R. Mazhar, M. A. Kadir,Z. B. Mahbub, K. R. Islam, M. S. Khan, A. Iqbal, N. Al-Emadi et al.,“Can AI help in screening viral and COVID-19 pneumonia?” arXivpreprint arXiv:2003.13145, 2020.

[70] H. Panwar, P. Gupta, M. K. Siddiqui, R. Morales-Menendez, andV. Singh, “Application of Deep Learning for fast detection of COVID-19 in x-rays using nCOVnet,” Chaos, Solitons & Fractals, p. 109944,2020.

[71] B. Hurt, S. Kligerman, and A. Hsiao, “Deep Learning localizationof pneumonia: 2019 coronavirus (COVID-19) outbreak,” Journal ofThoracic Imaging, vol. 35, no. 3, pp. W87–W89, 2020.

[72] D. Ezzat, H. A. Ella et al., “GSA-DenseNet121-COVID-19: a hybriddeep learning architecture for the diagnosis of COVID-19 diseasebased on gravitational search optimization algorithm,” arXiv preprintarXiv:2004.05084, 2020.

[73] S. Minaee, R. Kafieh, M. Sonka, S. Yazdani, and G. J. Soufi, “Deep-COVID: Predicting COVID-19 from chest x-ray images using deeptransfer learning,” arXiv preprint arXiv:2004.09363, 2020.

[74] D. Das, K. Santosh, and U. Pal, “Truncated inception net: COVID-19 outbreak screening using chest x-rays,” Physical and engineeringsciences in medicine, pp. 1–11, 2020.

[75] S. Basu and S. Mitra, “Deep Learning for screening COVID-19 usingchest x-ray images,” arXiv preprint arXiv:2004.10507, 2020.

[76] M. Rahimzadeh and A. Attar, “A new modified Deep ConvolutionalNeural Network for detecting COVID-19 from x-ray images,” arXivpreprint arXiv:2004.08052, 2020.

[77] Y. Zhang, S. Niu, Z. Qiu, Y. Wei, P. Zhao, J. Yao, J. Huang, Q. Wu, andM. Tan, “COVID-da: Deep domain adaptation from typical pneumoniato COVID-19,” arXiv preprint arXiv:2005.01577, 2020.

[78] I. D. Apostolopoulos, S. I. Aznaouridis, and M. A. Tzani, “Extract-ing possibly representative COVID-19 biomarkers from x-ray imageswith Deep Learning approach and image data related to pulmonarydiseases,” Journal of Medical and Biological Engineering, p. 1, 2020.

[79] H. Choi, X. Qi, S. H. Yoon, S. J. Park, K. H. Lee, J. Y. Kim, Y. K. Lee,H. Ko, K. H. Kim, C. M. Park et al., “Extension of coronavirus disease2019 (COVID-19) on chest ct and implications for chest radiographinterpretation,” Radiology: Cardiothoracic Imaging, vol. 2, no. 2, p.e200107, 2020.

[80] H. Shi, X. Han, N. Jiang, Y. Cao, O. Alwalid, J. Gu, Y. Fan, andC. Zheng, “Radiological findings from 81 patients with COVID-19 pneumonia in Wuhan, China: a descriptive study,” The LancetInfectious Diseases, 2020.

[81] A. A. Ardakani, A. R. Kanafi, U. R. Acharya, N. Khadem, andA. Mohammadi, “Application of Deep Learning technique to manageCOVID-19 in routine clinical practice using ct images: Results of 10convolutional neural networks,” Computers in Biology and Medicine,p. 103795, 2020.

[82] L. Li, L. Qin, Z. Xu, Y. Yin, X. Wang, B. Kong, J. Bai, Y. Lu, Z. Fang,Q. Song et al., “Artificial Intelligence distinguishes COVID-19 fromcommunity acquired pneumonia on chest ct,” Radiology, p. 200905,2020.

[83] F. Shan, Y. Gao, J. Wang, W. Shi, N. Shi, M. Han, Z. Xue, and Y. Shi,“Lung infection quantification of COVID-19 in ct images with DeepLearning,” arXiv preprint arXiv:2003.04655, 2020.

[84] A. Holzinger, M. Plass, M. Kickmeier-Rust, K. Holzinger, G. C.Crişan, C.-M. Pintea, and V. Palade, “Interactive machine learning:experimental evidence for the human in the algorithmic loop,” AppliedIntelligence, vol. 49, no. 7, pp. 2401–2414, 2019.

[85] Z. Tang, W. Zhao, X. Xie, Z. Zhong, F. Shi, J. Liu, and D. Shen, “Sever-ity assessment of coronavirus disease 2019 (COVID-19) using quanti-tative features from chest ct images,” arXiv preprint arXiv:2003.11988,2020.

[86] O. Gozes, M. Frid-Adar, H. Greenspan, P. D. Browning, H. Zhang,W. Ji, A. Bernheim, and E. Siegel, “Rapid ai development cycle forthe coronavirus (COVID-19) pandemic: Initial results for automateddetection & patient monitoring using Deep Learning ct image analysis,”arXiv preprint arXiv:2003.05037, 2020.

[87] J. Zhao, Y. Zhang, X. He, and P. Xie, “COVID-ct-dataset: a ct scandataset about COVID-19,” arXiv preprint arXiv:2003.13865, 2020.

[88] C. Butt, J. Gill, D. Chun, and B. A. Babu, “Deep Learning systemto screen coronavirus disease 2019 pneumonia,” Applied Intelligence,p. 1, 2020.

[89] S. Wang, B. Kang, J. Ma, X. Zeng, M. Xiao, J. Guo, M. Cai, J. Yang,Y. Li, X. Meng et al., “A Deep Learning algorithm using ct images toscreen for corona virus disease (COVID-19),” MedRxiv, 2020.

[90] X. Bai, C. Fang, Y. Zhou, S. Bai, Z. Liu, L. Xia, Q. Chen, Y. Xu,T. Xia, S. Gong et al., “Predicting COVID-19 malignant progressionwith AI techniques,” 2020.

[91] Y. Pathak, P. K. Shukla, A. Tiwari, S. Stalin, S. Singh, and P. K. Shukla,“Deep Transfer Learning based classification model for COVID-19disease,” IRBM, 2020.

[92] S. Jin, B. Wang, H. Xu, C. Luo, L. Wei, W. Zhao, X. Hou, W. Ma,Z. Xu, Z. Zheng et al., “Ai-assisted ct imaging analysis for COVID-19screening: Building and deploying a medical AI system in four weeks,”medRxiv, 2020.

[93] C. Jin, W. Chen, Y. Cao, Z. Xu, X. Zhang, L. Deng, C. Zheng, J. Zhou,H. Shi, and J. Feng, “Development and evaluation of an AI system forCOVID-19 diagnosis,” medRxiv, 2020.

[94] G. Wang, X. Liu, C. Li, Z. Xu, J. Ruan, H. Zhu, T. Meng, K. Li,N. Huang, and S. Zhang, “A noise-robust framework for automaticsegmentation of COVID-19 pneumonia lesions from ct images,” IEEETransactions on Medical Imaging, 2020.

[95] J. Song, H. Wang, Y. Liu, W. Wu, G. Dai, Z. Wu, P. Zhu, W. Zhang,K. W. Yeom, and K. Deng, “End-to-end automatic differentiation ofthe coronavirus disease 2019 (COVID-19) fr