arXiv:2111.09537v1 [cs.LG] 18 Nov 2021

The Prominence of Artificial Intelligence in COVID-19


MD Abdullah Al Nasim [email protected] of Research and DevelopmentPioneer Alpha

Aditi Dhali [email protected] of Research and DevelopmentPioneer Alpha

Faria Afrin [email protected] of Research and DevelopmentPioneer Alpha

Noshin Tasnim Zaman [email protected] of Research and DevelopmentPioneer Alpha

Nazmul karim [email protected] and Computer EngineeringUniversity of Central Florida

AbstractIn December 2019, a novel virus called COVID-19 had caused an enormous number ofcausalities to date. The battle with the novel Coronavirus is baffling and horrifying afterthe Spanish Flu 2019. While the front-line doctors and medical researchers have madesignificant progress in controlling the spread of the highly contiguous virus, technology hasalso proved its significance in the battle. Moreover, Artificial Intelligence has been adoptedin many medical applications to diagnose many diseases, even baffling experienced doctors.Therefore, this survey paper explores the methodologies proposed that can aid doctors andresearchers in early and inexpensive methods of diagnosis of the disease. Most developingcountries have difficulties carrying out tests using the conventional manner, but a significantway can be adopted with Machine and Deep Learning. On the other hand, the access todifferent types of medical images has motivated the researchers. As a result, a mammothnumber of techniques are proposed. This paper first details the background knowledge ofthe conventional methods in the Artificial Intelligence domain. Following that, we gatherthe commonly used datasets and their use cases to date. In addition, we also show thepercentage of researchers adopting Machine Learning over Deep Learning. Thus we providea thorough analysis of this scenario. Lastly, in the research challenges, we elaborate on theproblems faced in COVID-19 research, and we address the issues with our understandingto build a bright and healthy environment.Keywords: COVID-19, CT image, X-ray, Deep Learning, Machine Learning

1. Introduction

After the flu of 1918, the novel Coronavirus or COVID-19 became the fifth documentedoutbreak and was discovered first in Wuhan; China Li et al. (2020). Gradually, it spreadin the whole world. The International Committee on Taxonomy of Viruses gave the officialname of the Coronavirus as “Severe Acute Respiratory Syndrome Coronavirus 2” or SARS-

1

arX

iv:2

111.

0953

7v1

[cs

.LG

] 1

8 N

ov 2

021


Fig. 1: The Transmission Mechanism of COVID-19 from One Species to Another

CoV-2 Liu et al. (2020). Rai et al. (2021) mentioned that through respiration or in touch withinfected droplets during the incubation period (2-14 days). The World Health Organization(WHO) announced the SARS-CoV-2 outbreak as a Public Health Emergency of InternationalConcern on January 30, 2020, and a pandemic on March 11, 2020 Cucinotta and Vanelli(2020). It got reports of a cumulative total of 116,736,437 confirmed cases globally untilMarch 19, 2021, according to PAHO.

If we look into the history, Zhao et al. (2020), considered bats as the natural host ofSARS-CoV-2. Moreover, Platto et al. (2021) believed that bats are a locus for its expansion.According to Phan et al. (2020), Coronavirus infection which arrived from China incre-ased anxiety about human-to-human transmission. Riou and Althaus (2020) explained thathuman-to-human communication of 2019-nCoV is similar to the SARS-CoV of 2002. Theauthors suggested giving extra attention to the prevention of COVID-19. They showed thatit is possible to end the epidemic according to the SARS-CoV of 2002 experience. In thatcase, the screening and control system at transportation stations and airports may play avital role in preventing the outbreak of COVID-19. Fig. 1 shows the transmission of theCoronavirus.

From the ending of 2019, countries started to take precautions to prevent its outbreak. Atthat time, the Chinese government announced a lockdown for three weeks in Hubei provinceafter the explosion. It imposed restrictions on traveling, transportation, public gathering tomaintain social distance in Wuhan. Though China maintained a strict lockdown, the rateof positive COVID-19 cases and deaths increased significantly. Dechsupa et al. (2020) wroteabout the pandemic situation in Thailand. They said that its government declared a curfewon April 3, 2020, asking people to wear a face mask, maintain social distance and stay athome from 10 p.m. to 4 a.m. Barkur and Vibha (2020) referred that a one-day curfew named“Janata Curfew” was given in India. The Indian government later extended the curfew forsafety.

2


Fig. 2: The COVID-19 Data of the Pie Chart was Collected from John Hopkins UniversityCSSE GitHub at 2:30 am on 30th June 2021

Fig. 2 shows the percentage of Daily COVID-19 cases and Deaths on June 30, 2021,based on the top ten affected countries. It is cleared that the number of total daily cases anddeaths is more prominent in the United States than in the other nine countries. The secondmost affected country is India which is 19 percent, and the third one is Brazil receiving15 percent of total daily cases and deaths. The lowest percent of daily cases and deaths isin Germany with 3 percent. Among European countries, Italy was the first affected by theCoronavirus. Within March 18, 2020, the number of COVID-19 patients and deaths wereover 30,000 and 2500 Spinelli and Pellino (2020). The total number of affected cases, deathcases, and recovered cases around the world from the very beginning of this pandemic tillJuly 11, 2021, is depicted in Fig. 3. Affected cases, death cases, and recovered cases basedon different countries are shown in Fig. 4. We gathered these data from online (wor (2021),Holder (2021), pha (2021)). The United States of America tops this list with a total numberof affected cases (34,726,111) when total death cases are 622,821 and the recovery rate isapproximately 81.07%.

Medical centers and hospitals are more crowded nowadays as the number of virus-infectedpeople increases due to new triple mutants, and their respiratory condition worsens. So,it is tough to serve everyone at the same time. But, the medical science sector is workingrelentlessly on research on how to stop the spread of the virus, reduce human anxiety, increaseimmunity, make the proper vaccine for this infectious disease, and serve each patient.

Mental Health and COVID-19: Huang and Zhao (2021) showed the mental healthcondition of people for the outbreak of COVID-19 in China. They found out that peopleyounger than 35 years have a higher prevalence of anxiety symptoms, and those who aregreater than 35 years old have depressive symptoms. The authors suggested that there shouldbe continuous monitoring of the psychological impacts for the high-risk population as partof targeted interventions during COVID-19.

3


Fig. 3: Total Affected Cases, Death Cases and Recovered Cases Around the World until 11July, 2021

Cou

ntry

USAIndia

BrazilFranceTurkeyRussia

UKItaly

GermanySpain

ArgentinaBangladesh

0 20,000,000 40,000,000 60,000,000

Total case Total death Total recovered

Fig. 4: Total Affected Cases, Death Cases and Recovered Cases of Different Countries until11 July, 2021

COVID-19 Infection and Immune Response: Shi et al. (2020) mentioned incu-bation and non-severe stages where a particular adaptive immune response is needed fordestroying the SARS-CoV-2 and averting its progress to the following steps. They suggestedthat monitoring the patients who recover from the non-severe stage for the SARS-CoV-2with T/B cell response is necessary. It should be taken into consideration while deciding thestrategies of the vaccine according to them.

Vaccination: Mahase (2021) explained the vaccine efficiency of Novavax’s vaccine andJohnson & Johnson vaccine. Novavax is a US biotechnology company, and its vaccine’s ef-ficiency results in 95.6% against the original variant of SARS-CoV-2. It gives protectionagainst the newer variants B.1.1.7, which is 85.6%, and B.1.351 is 60%. In addition, a phaseIII trial of that vaccine was implemented in the UK with over 15,000 participants who arebetween 18 years old and 84 years old. The authors also mentioned that the Johnson &

4


Fig. 5: The Percentage of People Vaccinated with Respect to t heir Countries until 11 July,2021

Johnson vaccine has an efficiency of 66 percent of one dose. Johnson & Johnson vaccines arecapable of preventing moderate to severe illness, which lasts for 28 days after vaccination.The phase III trial of that vaccine was implemented on 43,783 participants of US (44%),Central and South America (41%), and South Africa in the United States (15%). The ful-ly vaccinated countries are illustrated in Fig. 5. UK leads here also having 51% of theirpopulation fully vaccinated (Holder (2021), pha (2021)).

Following the discovery of the COVID-19 pandemic, the importance of the models inovercoming COVID-19 challenges has multiplied. Taking these factors into consideration,in this paper, we have conceptualized and comprehensively represented a systematic reviewof the recent research on Machine Learning (ML) and Deep Learning (DL) in detectingCOVID-19. This paper aims to effectively identify the ML and DL algorithms that helpedin the early detection of COVID-19.

The whole work we have done in this paper is depicted in this Fig. 6. At first, welooked for research papers into famous online databases such as Elsevier, ERS journals,IEEE Xplore, Springer, Google scholar, Hindawi, MPDI, science direct, etc. On differentwebsites, we have searched with some common keywords regarding the detection of COVID-19 using ML and DL methods and gather more than 700 papers. A graphical representationof different publishers and the number of documents we have used from those to developthis work has been depicted in Fig. 7. We have filtered out more than 300 papers focusingon the documents that have been used for early detection or solving COVID-19 relatedissues. Exclusion procedures for filtering more than 250 publications that apply to COVID-19. Finally, we have reviewed a total of 100 papers related to ML and DL models in thedetection of COVID-19 and some non-ML and DL algorithms that use image processingand computer vision algorithms in the early detection of COVID-19 and other COVID-19related crises. In total, 100 papers, 35 papers used ML frameworks, 60 articles used DLalgorithms, and five papers on non-ML and non-DL algorithms. We have separated theminto different parts: methodology, summary, and the complete overview of other researchwork using tables. Data derived from each research paper evaluated for the analysis also

5


Fig. 6: Flowchart of Our Survey Work. It Demonstrates the Workflow from the Identificationof Research Aim till the Collection of the Prominent Researches over the RecentYears in the COVID-19 Research Community

include dataset used, the tool used, type of input used, augmentations technique, no. ofimages used, accuracy, and other measures are highlighted.

2. Background Study

Machine Learning is a field of computer science that allows software systems to learn auto-matically and improve from experience without any explicit programming. It can adapt tonew data through iterations independently. ML is divided into mainly three classes: supervi-sed learning, unsupervised learning, and reinforcement learning, where each one has specificactions and particular results.

6


Fig. 7: Plot to Demonstrate Our Collection of Research Papers from Different RenownedPublishers

2.1 Supervised Learning

Supervised learning is a method where an algorithm is trained on labeled input data for aparticular output. From the beginning of training dataset analysis, this learning algorithmis trained until it discovers underlying patterns and relationships between input and outputdata. It processes an inferred functions to yield accurate labeling results in the testingdataset. Sometimes, validation sets work as the control parameters, and the sets are part oftraining datasets. The purpose of supervised learning is to discover the best output withinthe context of a specific question. It has enough knowledge to compare the work with theintended output and error detection to adapt the model accordingly.

This method is divided into two types of problems: regression and classification. Regres-sion algorithms are used for the prediction of continuous variables. There are some practicalregression algorithms. For example Linear Regression, Regression Trees, Non-Linear Re-gression, Bayesian Linear Regression Polynomial Regression. A classification algorithm istypically used when output is split into two categories. The top algorithms of it are: Ran-dom Forest, Decision Trees, Logistic Regression, Support Vector Machines. ConvolutionalNeural Networks (CNN) by Al-Saffar et al. (2017); Edraki et al. (2020), Recurrent Neu-ral Network (RNN) by Sak et al. (2014), Long Short-Term Memory (LSTM) by Sak et al.(2014), Multilayer perceptron (MLP). These are the types of DL that are generally trainedas supervised methods. Both unsupervised and supervised learning procedures can preparethe Radial-Basis-Function Networks (RBFN).

2.2 Unsupervised Learning

Unsupervised learning, also known as unsupervised machine learning, is a technique to ana-lyze and cluster unlabeled datasets. The training and testing datasets are neither classifiednor labeled. It results in finding hidden patterns or data groupings without any human in-tervention. It infers a function that explores the data and builds inferences from datasets.It can detect similarities and varieties in the dataset to make an informed solution andrepresent it in a compressed format.

7


Unsupervised learning is used in clustering and Association problems. Clustering is usedto group objects where most similar things remain in a particular group. It finds the similarityin data and categorizes them. Association is a rule-based method used for searching for therelation among different variables from the gigantic database. It identifies features from thefrequency of the event. This type of algorithm is suitable for market analysis.

Some popular unsupervised learning algorithms: K-means clustering, KNN (k-nearestneighbours), Hierarchal clustering by Johnson (1967), Anomaly detection by Chandola et al.(2009), Neural Networks by Sarle (1994), Principle Component Analysis by Wold et al.(1987), Independent Component Analysis, Apriori algorithm, Singular value decomposition.Deep Belief Network (DBN) by Chen et al. (2015), Auto-Encoders by Rifai et al. (2011), DeepBoltzmann Machine (DBM) by Salakhutdinov and Hinton (2009), Generative AdversarialNetworks (GAN) by Mnih et al. (2013) are the types of unsupervised DL algorithms.

2.3 Reinforcement Learning

Reinforcement learning (RL) by Sutton and Barto (2018) is a rule-based technique thatdeals with the ability to learn the associations between stimuli, actions, and the occurrenceof events. It makes sequential decisions where an intelligent agent knows to behave in anenvironment by performing the activities and seeing the outcome of actions. RL has a widerange of applications in many fields.

2.4 Deep Learning

Deep learning [Goodfellow et al. (2016)] is a type of ML that allows a machine to mimichuman behavior. It is a subset of artificial intelligence. Deep convolutional nets have achievedleap forwards in handling pictures, images, video, and sound [LeCun et al. (2015)]. MLprepares machines for AI by training data and employing various algorithms, emphasizinglearning from experience.

The structure of organic neurons or human/animal brains encourages DL, allowing thecomputer to see, learn, and react like a person. It’s utilized for detecting objects, creatingpatterns, and identifying images or voices, among other things. The best examples are:Virtual assistants, self-driving cars, chatbots, face-speech recognition, self-driving cars, he-althcare applications. Other DL applications. Humans require a significant amount of timeto grasp large amounts of unstructured data and extract knowledge. However, if this vastvolume of data can be processed fast, any organization or sector can benefit. As a result,DL is driving huge advancements in transportation, medical, and other industries. The techindustry, in particular, is undergoing profound transformations as a result of the progressivedevelopment in deep understanding in several industries.

A stratified, deep artificial neural network is created by layering several artificial neuronstogether in a layered structure with most of the levels concealed. The number of layers canbe used to gauge its effectiveness. The more hidden stories in the network, the more “deep”the web will be, and the more complex non-linear linkages will be seen. A picture, speech,text, or another piece of data is provided as input, and a classification judgment is madebased on it. To tackle complex problems requires a large amount of data. It, on the otherhand, has tremendous computer capacity and can work with any data.

8


Fig. 8: The Common Methodology of Deep Learning in Training and Evaluating DifferentModels with COVID-19 Datasets

Deep learning is playing a big part in COVID 19 detection. A significant number of stra-tegies for identifying COVID 19 patients have already been developed by researchers. In Fig.8, we try to convey the core of the situation. We can state that we examine and follow thisflowchart in most of the paper to draw their Coronavirus prediction utilizing DL. Most ofthem use CT (Computed Tomography) or real-time reverse transcription-polymerase chainreaction (real-time RT-PCR) or CXR pictures of the lungs as input (Chest X-ray). ThoughCT scans are not always as accurate as RT-PCR, they are less expensive, more accessible,more scalable, making them an intriguing choice for researchers. However, RT-PCR has ahigh specificity, its sensitivity changes with time. The data is then preprocessed using va-rious data augmentation, normalization, and other procedures to improve the structure andproduce a vast dataset. Many of the authors cited a tiny dataset as one of their constraints.These methods can assist you in dealing with specific situations. Following that, severalfeatures are retrieved to use the given strategy ideally. Data is separated into training andtesting sets during the processing step. To carry out the tests, the authors utilized variousratios of training and testing datasets. They do the fine-tuning and cross-validation if theresult is not satisfactory enough. Here, as the model, some used the existing DL classifiers,such as ResNet, EfficientNet-B0, MobileNetV2, DenseNet, Inception V3, or SqueezeNet, etc.Some other authors have applied their proposed model here. For example, Covid-resnet Fa-rooq and Hafeez (2020), Covid-caps Afshar et al. (2020), MetaCOVID Shorfuzzaman andHossain (2021), Ensemble-CVDNet Öksüz et al. (2020), COVIDScreen Singh et al. (2021),DL-CRC Sakib et al. (2020), Covid-fact Heidarian et al. (2021) etc. As output, they revealthe prediction. Some predicted whether the input belongs to a COVID-19 patient or not. So-me identified COVID 19 patients from other pneumonia patients or virus-affected patients.Evaluation or validation is done by accuracy, AUROC, F1 score, AUC, or recall.

9


3. Datasets

Most scientific research depends on the collection and experiment of measurement data.Some datasets are private, and some are public for the authorities, politicians, researchers,and doctors. These scientific data sets are applied to develop new algorithms and differentiatethose from other algorithms in the particular study. Nowadays, some datasets such as medicalimage data, electronic health records, country-wise data, public data, etc., are being usedconstantly for scientific research.

3.1 Medical Image Data

Medical image data are gathered from X-rays, CT scans, MRI, intra-operative navigation,and biomedical research. It is the primary data source in medical science. Some renownedmedical image data sources for COVID-19: Cohen Dataset, Praveen: Corona Hack: Chest X-Ray Dataset, Mooney P. Chest X-ray Images (Pneumonia) Kaggle Repository, PneumoniaDataset, CORD-19 dataset Wang et al. (2020b).

3.2 Electronic Health Record

Electronic health records, or EHRs, are real-time data collecting systems that save patientinformation in a digital format and make it accessible to authorized users quickly and se-curely. The patient’s medical history, diagnosis, radiological images, and test results are allincluded. Medical errors are reduced, and patient data is protected, thanks to electronic he-alth records. The most commonly utilized EHR datasets include Research Electronic DataCapture (REDCap), JSRT Dataset (Japanese Society of Radiological Technology), Covid-19Radiography Database, and Italian Society of Medical and Interventional Radiology (SIRM)COVID-19 Database.

3.3 Country Data

The country dataset is the resource where users get different variables for specific coun-tries. Country data provides transparency and creates value to the community by makingit publicly available. Some authors used country data in the papers: Zoabi et al. (2021) de-veloped their model based on this. The Israeli Ministry of Health publicly disclosed datasetThe dataset contains primary records of all the residents who were tested for COVID-19countrywide. Elsheikh et al. (2021) collected the Covid-19 recovered cases, confirmed cases,and deaths from the Saudi Ministry of Health website. They used the dataset as time-seriesdata to train their proposed model. Moreover, Shastri et al. (2020) used the Ministry of He-alth and Welfare, India and Centers for Disease Control and Prevention, U.S. Departmentof Health and Human Services for their paper with Convolutional LSTM model.

USA based Covid-19 outbreak analysis was done by Haratian et al. (2020) and Bashiret al. (2020). Baqui et al. (2020), de Souza Santos et al. (2021), Cota et al. (2020), Ribeiroet al. (2020), Wollenstein-Betech et al. (2020), Silva and Figueiredo Filho (2021) analysedCovid-19 related data of Brazil. Bhatnagar et al. (2020), Yadav (2020), Ulahannan et al.(2020), Roy et al. (2021), Gupta and Pal (2020), Bagal et al. (2020) worked on IndianCovid-19 cases. Researchers studied bout country-wise characteristics, spreading rate, themortality rate of Covid-19 in their research. For example, Al-Qaness et al. (2020) considered

10


the cases of Italy, the USA, Iran and Korea. Zarikas et al. (2020) scrutinized the data from30 countries including the USA, Brazil, China, South Korea, Spain, Canada. Al-Qaness et al.(2021) worked on Russia and Brazil’s outbreak .

Tabela 1: Commonly Used Dataset for Covid-19 Research

SerialNo. Name Reference Citations Optimal

Techniques

1 CohenDataset

Cohen et al.(2020) 491

Local Patch-Based Method,Inception withFractional-Order,InstaCovNet19,VGG16, ResNet-34,17 DepthwiseSeparable Conv.layers, ResNet-50,ResNet-101,DenseNet-121,ResNet-152,LR and PCA,24-Layer CNN,SP-COVID-Net,Bayesian Conv.Neural Networks,Inceptionv3,COVID-CheXNet,EfficientNet-B0,LBP+SVM

2 - Wang et al.(2017) 1374

VGG16,ResNet-34,EfficientNet-B0,ResNet50

3 JSRTDataset

Shiraishiet al. (2000) 620

Local Patch-Based MethodOh et al. (2020),PolynomialRegressionPunn et al. (2020)

4 SCRDatabase

van Nede-rveen Me-erkerk(2006)

472Local Patch-Based MethodOh et al. (2020)

5 MCDataset

Jaeger et al.(2014) 311 Local Patch

-Based Method

11


Table 1 continued from previous pageSerialNo. Name Reference Citations Optimal

Techniques

6 JHU CSSERepository - -

KNN, SVM, DTs,Holt winterSaba et al. (2021),MLPCar et al. (2020)

7Covid-19RadiographyDatabase

Chowdhuryet al. (2020) 197

InstaCovNet19Saba et al. (2021),Inception withFractional-orderSahlol et al. (2020),24-Layer CNNNath et al. (2020),VGG16Waheed et al. (2020)

8 - Rajpurkaret al. (2017) 1064

Inception withFractional-orderSahlol et al. (2020)

9Corona Hack:Chest X-RayDataset

- -Local Patch-Based MethodOh et al. (2020)

10 PneumoniaDataset - 153

DenseNet-121Rasheed et al. (2021),LBP+SVMSethy et al. (2020),ResNet18Khalifa et al. (2020)

11 COVIDxDataset

Wang et al.(2020a) 661 COVID-Net

12Dr. AdrianRosebrock‚Äôs Dataset

Hemdanet al. (2020) - DenseNet201

Hemdan et al. (2020)

13

RSNAPneumoniaDetectionDataset

Wang et al.(2020c) -

ResNet-101,ResNet-152,DenseNet201,Resnet50-V2,Inception-v3

14ChestImagingon Twitter

Das et al.(2021) -

DenseNet201,Resnet50-V2,Inception-v3

12


Table 1 continued from previous pageSerialNo. Name Reference Citations Optimal

Techniques

15

COVID-19Chest X-rayDatasetInitiative

Das et al.(2021) -

DenseNet201,Resnet50-V2,Inception-v3

Some other datasets are COVIDiSTRESS Yamada et al. (2021) that analyze behavioralconsequences of Covid-19, Geocov19 note qazi2020geocov19 that consists of a vast numberof multilingual Covid-19 related tweets. Melo et al. also came up with the first public datasetfrom Brazilian twitter de Melo and Figueiredo (2020). This dataset extends the Covid-19research from a different perspective. Some popular datasets used in various research worksrelated to COVID-19 are referenced in Table. 1 and the usage of these datasets in some mostcited papers are also shown. We are now about to discuss the details and object of someresearch works where ML detects Coronavirus-affected patients.

4. Deep Learning in COVID-19 Battle

Machine learning has begun to shine a light on hope in the face of the Covid-19 pandemicin this unclear and dire circumstance. ImageNet Classification, a CNN-based DL solution,earned significant success in the best-known international computer vision competition Su-zuki (2017), which sparked the adoption of ML in medical imaging area. Experts utilise MLto anticipate the unknown future by predicting who is most likely to be affected, the dangerof infection, diagnosing patients, developing treatments faster, and many other things. As aresult, it’s much easier to fight a conflict you know is coming rather than a large number ofadversaries that appear out of nowhere. That’s where Ml comes in, by forecasting the nextpandemic, treatment outcomes, and disease propagation.

4.1 Artificial Neural Networks In Covid-19

Wang et al. (2020c) created a Covid-19 diagnostic DL model that was utilized on chestradiographs and categorized the X-ray images using transfer learning and combination-basedCOVID-Net. They developed ResNet-101 and ResNet-152 and trained on the ImageNetdataset according to the precision and loss rate after correcting and integrating the transferlearning model with a modified deep network layer. Because this is an X-ray-based diagnosis,contrast enhancement was used as a preprocessing step. The multilayer perceptron artificialneural network structure was used by Borghi et al. (2021) to effectively describes and predictthe number of infected cases and deaths up to six days. Their model was trained with thedata of 30 countries. Rathee et al. (2021) has categorized the status of COVID-19 patients’health into infected, uninfected, exposed, and susceptible. They have used Bayesian andbackpropagation algorithms.

Nath et al. (2020) used a Deep Neural Network to identify Covid-19 from x-ray pic-tures and CT scans using binary and multi-class classification in a study. Their proposedmodel is a 24-layer 6 CNN Network for X-ray and RT-PCR categorization. The ImageNetdataset suggested that CNN architecture outperforms AlexNet, DenseNet121, InceptionV3,

13


ResNet18, and GoogleNet regarding X-ray image classification accuracy. In the study ofMaiyana et al. (2021), they forecast the Covid-19 result based on the home industry usingneural networks with the backpropagation method.

Narin et al. (2020) used a deep transfer learning-based strategy to predict Covid-19 pa-tients using chest X-rays from four categories (normal, COVID-19, bacterial, and viral pneu-monia). They used pre-trained transfer models based on deep convolution neural networks,which resulted in a high-accuracy decision assistance system for radiologists. In addition, theproposed models include a complete framework that eliminates the need for manual featureextraction, selection, and categorization. They employed five distinct CNN pre-trained mo-dels (ResNet50, ResNet101, ResNet152, InceptionV3, and Inception-ResNetV2) to improveprediction accuracy for three binary datasets. They performed binary classifications usingfour classes (COVID-19, normal, viral pneumonia, and bacterial pneumonia). This researchused a variety of pre-trained models. A five-fold cross-validation procedure was applied toobtain a robust result. 80% of the data was set aside for training, while the remaining 20%was set aside for testing. With 96.1 percent, ResNet50 and ResNet101 had the best overallperformance. These two models’ accuracy and loss numbers are superior to those of the othermodels. COVID-Net, a deep convolutional neural network design for detecting COVID-19instances from chest X-ray (CXR) pictures, was introduced by Wang et al. (2020a). COVIDxis an open-access benchmark dataset that contains 13,975 CXR images from 13,870 patientcases, with the largest number of publicly available COVID-19 positive cases. It was usedto assess COVID-19 prediction. Fig. 17 is collected that shows the framework.

Zhang et al. (2020c) employed a two-step transfer learning pipeline/strategy and thedeep neural network architecture COVID19XrayNet to detect COVID-19 with only 189annotated COVID-19 X-ray images. This COVID19XrayNet framework is created using amodified ResNet34 structure that outperforms the original ResNet34 design. Gozes et al.(2020) wanted to build an AI-based detection tool that can measure and track Coronaviruswith high accuracy through CT image processing. This method determined if an image waspositive or negative for COVID. If any positive cases are discovered, the system generatesa localization map and lung measures. They created several components and divided theCT image cases into two levels. Subsystem A uses existing methods for the 3D analysis ofthe case volume for focal opacities and nodules. On the contrary, subsystem B uses newlydeveloped techniques for 2D analysis of every slice to localize and identify larger-sized diffuseopacities and ground-glass infiltrates, which is an essential indication for Coronavirus. Toisolate the region of interest in the lung, they used a lung segmentation module. They usedCT slices of abnormal lungs to train the U-net architecture for image segmentation. Theirregularities were detected using ResNet-50, a 50-layer deep neural network. The device pro-duced quantitative opacity measurements and a visualization of the more excellent opacityin a slice-based “heat map” or a 3D volume display for Coronavirus patients. A suggested“Corona score” is used to track how patients progress over time.

Ghoshal and Tucker (2020) attempted to determine how Bayesian Convolutional NeuralNetworks (Dropweights based) Networks (BCNN) can calculate uncertainty in DL resolu-tions to improve the diagnostic accuracy of the human-machine combination using COVID-19 chest X-ray in another paper. Their primary goal is to avoid false-negative detection. Weemployed a transfer learning method to build a Bayesian DL classifier using a pre-trainedResNet50V2 model to assess model uncertainty. They evaluate two variational-drop weights-

14


based uncertainty measures for image classification: Predictive Entropy (PH) and BayesianActive Learning by Disagreement (BALD). Furthermore, they employed the 1-Wassersteindistance to calculate uncertainty’s association with genuine error.

4.2 Long Short-Term Memory Networks in COVID-19

Shi et al. (2021) proposed a DL-based quantitative CT model that incorporated five clinico-radiological factors. They considered it a more efficient tool than PSI and individual qu-antitative CT parameters for predicting the severity of COVID-19 patients. The selectionoperator LASSO logistic regression analysis is used to characterize the baseline of clinical-radiological features. In one study, Abbasimehr and Paki (2021) proposed three hybrid me-thods for predicting the number of COVID-19 daily infected cases by combining differentDL models, such as CNN-based method, multi-head attention-based method, and LSTM-based method with the Bayesian Optimization Algorithm. Using laboratory data, Alakusand Turkoglu (2020) developed a clinical predictive model to determine whether a patientcould have Covid-19 illness. For model evaluation, they calculated precision, F1-score, re-call, AUC, and accuracy ratings. Then, they put the model to the test using data from 600patients in 18 labs. They next used 10-fold cross-validation and train-test split techniquesto confirm it. They looked at samples from Israel’s Albert Einstein Hospital. The authorstrained six models: ANN, CNN, LSTM, RNN, CNNLSTM, CNNRNN, and two hybrid mo-dels: CNNLSTM and CNNRNN. They used a trial-and-error method to set the parametersof each model. Elsheikh et al. (2021) proposed the LSTM network as a robust DL model.They used the publicly available data to train the model. Root mean square error (RMSE),coefficient of determination, mean absolute error (MAE), efficiency coefficient (EC), overallindex (OI), coefficient of variation (COV), and coefficient of residual mass were used toevaluate the model (CRM). They used their proposed approach to forecast the number ofconfirmed cases and deaths (Brazil, India, Saudi Arabia, South Africa, Spain, and the USA).

Devaraj et al. (2021) attempted to compare prediction models such as ARIMA, LSTM,Stacked LSTM, and Prophet techniques to forecast the spread of COVID-19 in terms ofinfected individuals, deaths, and recovered cases. They compared the forecasts’ accuracyand picked the best model based on the various performance measurements and statisticalhypothesis analysis. The procedure entailed putting the models to the test by using themto simulate global growth scenarios. For 32 states and union territories in India, Jamshidiet al. (2020) utilized DL models to forecast the number of Coronavirus positive reportedcases. To forecast COVID-19 positive cases, they used RNN-based LSTM variations suchas Deep LSTM, Convolutional LSTM, and Bi-directorial LSTM. The Convolutional LSTMmodel outperformed the other two models in the Shastri et al. (2020) study, with error ratesranging from 2.0 to 3.3 percent for all four datasets. With real-time COVID-19 data, therate of reliability of their best-proposed model, which offered the state-of-the-art outcome, issatisfactory. In their research, they looked into COVID-19 confirmed and fatality instances.

Table. 2 reflects some significant points of our work related to ML in COVID-19 patientdetection.

15


Tabela 2: Details of Papers that Accommodated Shallow Ma-chine Learning Techniques

References InputType

MethodTechniqueUsed

NumberofImagesData

Augmen-tationTechnique

CrossValida-tion

Callejon-Leblic et al.(2021)

Textdata

Compre-hensivemachinelearningmodelling

777patients NA 5 fold

Zoabi et al.(2021)

Textdata

Gradient-boostingmachinemodel

51,831indivi-duals

NA NA

Yao et al.(2020)

Clinicaldata

SVM, LR,RF, KNN,AdaBoost

137cases NA NA

Barstuganet al. (2020)

CTimages

SupportVectorMachine

150CTimages

NA 10 fold

Chen et al.(2021)

CTimages

SVM,FeatureExtractionandClassifi-cation

326 NA NA

Zeng et al.(2021)

Textdata

NLP,XGBoost,MLP,Bi-LSTM,SVR,Grid-SearchCV

301,363patients NA NA

Rasheedet al. (2021)

X-rayimages LR, PCA 500

images Scaling NA

Elaziz et al.(2020)

X-rayimages

FrMEM,MRFODE,KNNclassifier

2451images NA NA

Zhang et al.(2020b)

X-rayimages

EfficientNet,ConfidNet,AnoDet

519images NA 5 fold

16


Table 2 continued from previous page


MethodTechniqueUsed

NumberofImagesData


CrossValida-tion

Cohen et al.(2020)

X-rayimages NA 679

images NA NA

Loey et al.(2021)

Imagedata

SupportVectorMachine,Resnet50,Ensemblealgorithm

26140images NA NA

Mele andMagazzino(2021)

Timeseriesdata

Toda-Yamamototest, MLanalysiswith D2Calgorithm

annualdata NA NA

Kassaniet al. (2020)

ChestX-ray,CTimages

ImageNet,VGGNet,NASNet,NASNetMobile,ResNet50,ResNet101V2,MobileNet,InceptionV3,Xception,DT, RF,XGBoost,AdaBoost,Baggingclassierand LightGBM

254images NA 10 fold

17




MethodTechniqueUsed

NumberofImagesData


CrossValida-tion

Sethy et al.(2020)

Xrayimages

SVM, CNN,ResNet50,HOG, GLCM,AlexNet,VGG16,VGG19,GoogleNet,ResNet18,ResNet50,ResNet101,InceptionV3,InceptionResNetV2,DenseNet201,andShuffleNet

381 NA NA

Khandayet al. (2020)

Textdata

TF/IDF,BOW,LogisticregressionandMultinomialNaı¨veBayes,decisiontree,randomforest,bagging,Adaboostandstochasticgradientboosting

212patients

Preprocessing(Punctuationandlemmatisation)

10 fold

18




MethodTechniqueUsed

NumberofImagesData


CrossValida-tion

Sujath et al.(2020)

Timeseriesdata

Linearregression,MultilayerperceptronandVectorautoregressionmethod

NotDefined NA NA

Patel et al.(2021)

Demo-graphicdata,clinicaldata,bloodpanelprofile

Randomforest,multilayerperceptron,supportvectormachines,gradientboosting,extra treeclassifier,Adaboost

— NA 5 fold

Tan et al.(2020)

Clinicaldata,CTimages

Auto-MLmethodofradiomics

319ofpatients

NA NA

Saba et al.(2021)

Timeseriesdata

ARIMA,SARIMA,GBR, RF,KNN, SVM,DTs, polynomialregression,Holt winter

NotMentioned NA NA

Elhadadet al. (2020)

Textdata

DT, KNN,LR, LSVM,MNB, BNB,Perceptron,NN, ERF,andXGBoost

7,486instances

Informationfiltering,TextParsing,DataCleaning

5 fold

19




MethodTechniqueUsed

NumberofImagesData


CrossValida-tion

Wang et al.(2020e)

CTimages

AlexNet,DeCoVNet,pretrainedUNet

630CTimages

Rando-maffinetrans-formation,colorjittering

NA

Car et al.(2020)

Timeseriesdata

MLP20706datapoints

Conversionof data 5 fold

Yang et al.(2020)

CTimage

ResidualU-Net,CSSL,Stochasticgradientdescent

812CTimages

cropping,colorjittering,Gaussianblur,randomgray-scaleconver-sion

NA

Zoabi et al.(2021)

RT-PCRresult

Gradient-boostingMachinemodel

99,232individuals

NA NA

Pourhomayounand Shakibi(2021)

Textdata

SupportVectorMachine,ArtificialNeuralNetworks,RandomForest,DecisionTree,LogisticRegression,andK-NearestNeighbor

2,670,000cases NA 10 fold

20




MethodTechniqueUsed

NumberofImagesData


CrossValida-tion

Heldt et al.(2021) EHR

Multivariatelogisticregression,randomforest,extremegradientboostedtree


Fernandeset al. (2021) EHR

Artificialneuralnetworks,randomforests,catboost,andextremegradientboosting


Shi et al.(2021) CT image

LASSO,logisticregressionanalysis,PredictiveNomogramMode


4.3 Convolutional Neural Networks in COVID-19

We categorize this section into three main parts. They are extensive scale networks, medium-scale networks, and small-scale networks. Each of them is again described in some subsec-tions.

4.3.1 Large Scale Network in COVID-19

Due to the COVID-19 epidemic, compiling a significant dataset to train a deep neuralnetwork has proven problematic. As a result, Oh et al. (2020) suggested a patch-based co-nvolutional neural network strategy suited for both training and evaluating the model usinga small training dataset for Covid-19 diagnosis. They presented a neural network design for

21


constructing this model. To begin, they preprocessed CXR images for data normalizationbefore extracting the lung sections. Second, they used the simple Restnet-18 as the backboneof a classification algorithm to classify the Chest X-Ray images according to their associa-ted diseases. The global and local approaches to implementing the categorization networkare the two alternative variations, with the local-patch-based process being their proposedmethod. The final decision is then made based on the majority vote. Finally, they employedthe GRAD-CAM saliency map to explain their technique. They presented a neural networkdesign for constructing this model. To begin, they preprocessed CXR images for data norma-lization before extracting the lung sections. After that, they employed FC-DenseNet103 fortraining purposes, employing the preprocessed data for segmentation network architecture.Second, they used the simple Restnet-18 as the backbone of a classification algorithm toclassify the Chest X-Ray images according to their associated diseases. The global and localapproaches to implementing the categorization network are the two alternative variations,with the local-patch-based process being their proposed method. The final decision is thenmade based on the majority vote. Finally, they employed the GRAD-CAM saliency map toexplain their technique.

Salman et al. (2020) sought to develop a DL model for detecting COVID-19 pneumo-nia on high-resolution X-rays, alleviating radiologists’ workload and contributing to theepidemic’s control. They define the proposed solution as selected convolutional network ar-chitecture and examine the design decisions, evaluation methodologies, and implementationissues that go along with it. They employed deep feature extraction based on deep learningarchitectures with their dataset.

Using CT Scan Datasets and Deep Convolutional Neural Networks:Yoo et al. (2020) propose a deep learning-based decision-tree classifier for quickly reco-

gnizing COVID-19 from CXR pictures, which may be utilized to make quick decisions aboutCovid patients. Three binary decision trees are used in their suggested classifier. The CXRpictures are classified as normal or abnormal in the first decision tree. The following treedifferentiates aberrant photos with tuberculosis indications, while the third does the samefor COVID-19. Ismael and Şengür (2021) presented a deep learning-based algorithm to di-stinguish between healthy and Covid-19 chest X-ray pictures in their work. Deep featureextraction, fine-tuning of pre-trained convolutional neural networks (CNN), and end-to-endtraining of a built CNN model were all used in the research.

Both CT Scans and X-rays Dataset:In a paper, Hussain et al. (2021) have proposed a novel CNN model which is based

on 22-layer CNN architecture for the automatic detection of COVID-19 as well as to findout the accurate diagnosis of two (COVID and Normal), three (COVID, Normal, and non-COVID pneumonia) and four (COVID, Normal, non-COVID viral pneumonia, and non-COVID bacterial pneumonia) class classification based on CT scan images and chest X-rayimages. Here, they have used Batch Normalization and, to assess the performance, usedfive-fold cross-validation.

The main objective of the paper of Pham (2021) was to find out the effectiveness ofpre-trained CNN models for the detection of COVID-19 for rapid development. Thus, theclassification tasks were fine-tuned using three public chest X-ray datasets and three pre-trained CNN models, namely SqueezeNet, AlexNet, and GoogleNet, without any data au-gmentation techniques. SqueezeNet and AlexNet, out of the three networks employed, take

22


the least time to train and predict. As a result, fine-tuned pre-trained models can achievemore incredible performance without using data augmentation with the proper selection oftraining parameters.

Two separate datasets, such as X-ray images and CT images, are utilized to validate themodels, with two distinct categories: COVID-19 and normal. Hussain et al. (2021) proposedthe CoroDet, a 22 layer CNN model that outperforms the two, three, and four class classifica-tion situations. Besides, the suggested model has higher accuracy than other state-of-the-artapproaches, indicating that it outperforms previous models. As the number of cases rises,the authors recommend this methodology to address the rising need for COVID-19 testingkits. As an alternative, they advise using CT scans and X-ray pictures with a DL model.

Using CT Scan Datasets and Xception:After collecting datasets from available resources, picture preprocessing was done utili-

zing data augmentation techniques such as horizontal flip, reshape image, zoom range, andrandom rotation range in another article by Jain et al. (2021). After that, three CNN mo-dels are employed to examine and compare their performance: ResNet, Inception V3, andXception. Instead of using the original ReLU activation function, LeakyReLU activation isused in the given models to speed up the training process. When it comes to detecting theCOVID-19, Xception outperforms all three models. Rahimzadeh et al. (2021) created an on-line automated method that could detect the Covid-19 from lung CT scans. The researchersreleased a new dataset of 63,849 CT scan images, 48,260 of which are normal and 15,589of which are infectious. They created a thorough model and tested it against other modelssuch as Xception and ResNet50V2. Finally, they achieved a 98.49 percent accuracy on 7996test samples using their unique approach.

Using X-rays Image Datasets and EfficientNet:To improve existing COVID19 detection using a limited number of publicly available

CXR datasets, Zebin and Rezvy (2021) designed and developed a CXR-based COVID-19disease detection and classification pipeline using a modified VGG-16, ResNet50, and newEfficientNetB0 architecture in this study. Moreover, they created a generative adversarialarchitecture to generate synthetic COVID-19 images during their studies to improve theunder-represented COVID-19 class. The lightweight structure of the convolutional backboneand their performance measurements in terms of accuracy, precision, and recall for correctlyrecognizing COVID-19 were essential factors in their selection for this study.

4.3.2 Medium Scale Network

Gupta et al. (2021) proposed a DL-based Computer-aided diagnostic Framework (InstaCovNet-19) to diagnose COVID-19 accurately and with lower misclassification rates. This InstaCovNet-19 model is a DL algorithm built from various pre-trained models, including ResNet-101,Inception v3, MobileNet, and NASNet. This CAD system fine-tunes pre-trained DL modelsfor feature extraction in COVID-19 patients’ chest X-rays. The SPP-COVID-Net technique,presented by Abdani et al. (2020), is the second most lightweight DL model appropriatefor a speedier screening process of the COVID-19 disease using mobile phone devices. Thisdeep network is built on the foundation of DarkCovid-Net, which has 14 layers of CNN.However, it is implemented by substituting the last few levels of the original network andparallel pooling layers of one SPP module.

23


Using X-rays Image Datasets and MobileNetV2:Using an x-ray of the chest To detect covid-19, Car et al. (2020) established a three-stage

model. After preprocessing, they employed five state-of-the-art transfer learning models,AlexNet, MobileNetv2, ShuffleNet, SqueezeNet, and Xception, along with three optimizers,Adam, SGDM, and RMSProp, to reduce over-fitting. Finally, they have classified the data.

Li et al. (2020) have developed a lightweight mobile app based on a deep neural ne-twork (DNN) that is used for screening and radiological trajectory prediction for COVID-19using chest X-ray. Besides, they depicted a novel three-player knowledge transfer and di-stillation (KTD) framework together with a pre-trained attending physician (AP) networkfor extracting image features from lung disease CXR images, a fine-tuned resident fellow(RF) network for learning the necessary CXR imaging features to differentiate betweenCOVID-19 and pneumonia or normal cases and a trained medical student (MS) network(e.g., MobileNetV2, or SqueezeNet) for operating COVID-19 patient triage and follow-upusing knowledge distillation. In addition, they apply novel loss functions and training codi-fication for the MS network to deal with the challenge of enormously similar and prevailingfore and background in medical images.

Using CT Scan Datasets and MobileNet:For the screening of COVID-19 cases, Sethi et al. (2020) suggested a model containing

four different pre-trained models: MobileNet, Xception, Inception V3, and ResNet50. Withan accuracy of 0.986, MobileNet outperformed the other competitor models. As a result,their work will be helpful as a diagnosis recommender system when the diagnostic burden ishigh. For the quick detection of COVID-19, Panahi et al. (2021) suggested a novel methodo-logy called FCOD, which was based on the Inception Network. They employed 17 DepthwiseSeparable Convolution Layers in this study, and their automatic detection system can reco-gnize complicated patterns from 2D X-ray images with a short forecast time. Furthermore,this automated detection system can recognize complex patterns from 2D X-ray images dueto its short prediction time. The methodology of Barbosa Jr et al. (2020) supports volume-tric percentage of airspace disease (consisting of ground-glass opacities and consolidations)on CT in COVID-19 patients for a better evaluation of the actual error when executing ADquantification on CXR. The ground truth volumetric quantification of COVID-19 relateddisease was initiated by manual segmentation on CT. Then, resulting 3D masks were pro-jected into 2D anterior-posterior radiographs (DRR) to measure area-based AD percentage(POa).

Using X-rays Image Datasets and VGG16:Nayak et al. (2021) employed eight pre-trained CNN models to build their proposed tech-

nique for quicker screening of COVID-19. With an accuracy of 98.33%, ResNet-34 outper-formed the other eight pre-trained models. For the speedier screening of COVID-19, ResNet-34, Inception-V3, GoogleNet, AlexNet, ResNet-50, MobileNet-V2, VGG-16, and SqueezeNetusing the notion of Transfer Learning were used. In addition, they used normalization anddata augmentation techniques to preprocess the gathered photos further. ResNet-34 out-performed the other eight pre-trained models with an accuracy of 98.33%. Furthermore,they have primarily concentrated on the data imbalances, attempted to overcome them, andpresented a reliable, automated, and cost-effective DL system with minimal resources.

Brunese et al. (2020) used deep learning (transfer learning) to detect the presence ofCOVID-19 in a chest X-ray. They used the VGG-16 model, which is a 16-layer convolutional

24


neural network. Additionally, they assured the explainability of the prediction by showingnetwork activation layers of the models. They had built two models and done this in threephases. Firstly, a thatting patient affected by a pulmonary disease from a healthy patient.The second step is to determine whether the lung condition is pneumonia or COVID-19.Finally, on the chest X-ray, identifying the areas with COVID-19 symptoms. For fine-tuning,the authors’ employed weights pre-trained on ImageNet on the VGG16 network. They cre-ated a new fully-connected layer head consisting of AveragePooling2D, Flatten, Dense, Dro-pout, and a final Dense layer on top of VGG16 to forecast the classes. They completed theirtuning by freezing the convolutional weights of VGG16. They employed the Grad-CAMtechnique for visual debugging. Finally, the performance of the classifiers was assessed usingfour metrics (Sensitivity, Specificity, F-Measure, and Accuracy).

Using X-rays Image Datasets and VGG19:

Sahlol et al. (2020) proposed a hybrid classification system dubbed Inception Fractional-order Marine Predators Algorithm, which is a blend of both Inception and Fractional-orderMarine Predators Algorithm (FO-MPA) (IFM). This approach was tested using two publiclyavailable datasets that included positive and negative chest X-Ray scan images of Covid-19patients. This classification strategy is used to classify the Covid-19 X-Ray pictures from rawimages without segmentation or preprocessing. As a result, input photos were first subjectedto a sophisticated CNN architecture to extract features. Then the FO-MPA method was usedto choose the most relevant features from the input images. This proposed combination isthe optimum strategy for removing and selecting deep features for image classification tasks.This proposed method successfully extracted and selected features 136 for dataset one and86 for dataset 2 out of 51K features with high accuracy.

Heidari et al. (2021) developed a novel model incorporating preprocessing steps for thedetection of COVID-19 pneumonia infected cases in their work and provided a quick decision-making tool for radiologists using computer-aided detection and diagnosis (CAD) systems.The VGG16 transfer learning-based CNN model is also employed for classifying chest X-raypictures, along with preprocessing methods.

Using a Small Dataset, Hall et al. (2020) attempted to investigate the use of chest X-Rayimages and how they might be used to detect the Covid-19 disease. They employed transferlearning without modifying any convolutional layers using two pre-trained Deep CNNs,VGG16 and Resnet50. The last layers of Resnet50 and VGG16 were deleted from their DLmodel, and the trainable section was replaced. The only augmentation technique they usedwas horizontal flipping. Later, they conducted a brief experiment using three classifiers onunseen Covid-19 (33 cases) and pneumonia (218 instances) points. Ten-fold cross-validationwas utilized to acquire the best estimate of the feasibility of employing chest X-Ray.

Song et al. (2021) introduced a new deep learning framework for autonomously diagno-sing COVID-19 patients from X-ray pictures based on seven different DCNN architectures:VGG19, DenseNet201, InceptionV3, ResNetV2, InceptionResNetV2, Xception, and Mobi-leNetV2. They suggested these for the best outcomes with the VGG19 and DenseNet201models. They used CT scans to build a speedy (in 30 seconds), accurate, and deep learning-based detection of COVID-19 patients, which can spontaneously extract the ground-glassopacity (GGO) from radiographs labeling the lesions from CT images. This model is alsocapable of distinguishing between individuals infected with COVID-19 and those infected

25


with bacteria pneumonia. Their achievement on Tianhe-2 supercomputers allowed for simul-taneous concurrent execution of thousands of tasks.

Using CT Scan Datasets and VGG19:Using CT scan pictures, Shah et al. (2021) used a variety of deep learning algorithms

to distinguish between COVID-19 and non-COVID 19. A self-made model named CTnet-10was built for the COVID-19 diagnostic, with an accuracy of 82.1 percent. To improve accu-racy, they passed the CT scan image through multiple pre-existing models. They discoveredthat the VGG-19 model, with a 94.52 percent accuracy, is the most accurate at identifyingideas as COVID-19 positive or negative. Hemdan et al. (2020) suggested a framework to as-sist radiologists in identifying Covid-19 patients. The COVIDX-Net includes seven differentdeep convolutional neural network models, including the modified Visual Geometry GroupNetwork (VGG19) and Google MobileNet’s second version. Each of them can deconstructthe standardized intensities of the X-ray image to determine if the patient is a COVID-19case that is positive or negative.

Waheed et al. (2020) developed an Auxiliary Classier Generative Adversarial Network(ACGAN) based model to generate synthetic X-ray (CXR) pictures to improve the perfor-mance of CNN-based COVID-19 detection. Besides, it will radiograph enhance the trainingdataset with the CNN model to improve detection efficiency. Panwar et al. (2020a) presenteda novel algorithm that can assist radiologists in the quick detection of COVID-19 with highaccuracy and low detection time using three datasets of Chest X-ray and CT-scan images.For the binary image classification of COVID-19 and Non-COVID-19 positive patients, theyused the VGG-19 transfer learning algorithm with five extra layers. Moreover, they useda Grad-CAM-based color visualization tool in their trial to interpret better and identifyradiological pictures.

Other Datasets and VGG:Gianchandani et al. (2020) proposed an ensemble model for the identification of COVID-

19 from infected, normal, and pneumonia people using two different transfer learning models,VGG16 and DenseNet. Data augmentation techniques such as vertical flipping, horizontalflipping, sheer transformation slant angle (0.2), and rotation by 45° are employed to pro-vide the best generalization capabilities. Following that, they tested the suggested modelusing two well-known datasets. Qjidaa et al. (2020) presented a three-phase clinical deci-sion support system based on chest radiography imaging for early diagnosis of COVID-19.First, afterradiographAgain, the preprocessing and feature extraction, a fully connected lay-er comprising many classifiers, is used to classify and predict. The VGG16 network is usedto create this model. Next, augmentation techniques like shearing, random flipping, and sca-ling preprocess the input photos. The next step is feature extraction, which is followed bytransfer learning. Finally, a fully linked layer of many classifiers is used to do classificationand prediction. The VGG16 network is used to create this model.

For the early detection of COVID-19, Qjidaa et al. (2020) have proposed a clinicaldecision support system that consists of three phases based on the chest radiographic images.After the preprocessing and feature extraction, classification and prediction are made witha fully connected layer of several classifiers. Here, the VGG16 network is chosen to form thismodel. For the preprocessing of input images, augmentation techniques are applied, namelyshearing, random flipping, and scaling. Secondly, feature extraction is followed by the transfer

26


learning step. Thirdly, classification and prediction are made with a fully connected layer ofseveral classifiers. Here, the VGG16 network is chosen to form this model.

Using CT Scan Datasets and GoogleNet:Zhou et al. (2021) provide an article for identifying Covid-19 using three popular deep

learning models named GoogleNet, ResNet, and AlexNet with transfer learning technique.They used preprocessed lung CT images to create 2500 high-quality pictures for training andvalidation. While DL models were utilized for feature extraction, majority voting proposedan ensemble classifier, EDL-COVID. They discovered that the ensemble method producedcomparable results in the detection of new coronaviruses. Using the ResNet model, Songet al. (2021) established a computer-aided technique for detecting Covid-19. The systemwas trained on 88 individuals with Covid-19 who had chest CT pictures and was able toidentify them. In addition, their program was able to detect lesion characteristics (Groundglass opacity), which will be extremely useful to frontline doctors. With an AUC score of0.95, the model could discriminate COVID-19 patients from healthy controls.

4.3.3 Small Scale Network

Using X-rays Image Datasets and DenseNet121:Wang et al. (2020d) presented a completely automatic innovative DL method to dia-

gnose Covid-19 and perform a prognosis analysis using bare chest pictures without the useof human annotation in another publication. The planned DL system is divided into threesections. First, they employed a completely automatic Densenet121-FPN DL model for au-tonomous lung segmentation to avoid non-lung areas. Second, a non-lung area suppressionoperation was proposed inside the lung-ROI, with supplemental methods S4 employed tosuppress non-lung area intensities. Then, to standardize the procedure, scaling and z-scorenormalization were applied. Finally, they developed an auxiliary training procedure for pre-training to allow the COVID-19Net to learn lung features from a vast dataset using aDenseNet-Like structure.

Using CT Scan Datasets and DenseNet:Kamal et al. (2021) in a paper, tried to evaluate and made compare all the eight pre-

trained models namely InceptionV3, NasNetMobile, ResNet50V2, DenseNet121, VGG19,ResNet50, MobileNet, MobileNetV2 for the efficient classification of chest X-ray images.DenseNet121 beats all other models to effectively classify Normal, COVID-19, SARS, Bac-terial Pneumonia, and Viral Pneumonia based on chest X-ray pictures. Panwar et al. (2020b)proposed a method called nCOVnet, a DL neural network-based fast screening (in 5 seconds)approach for identifying COVID-19 by analyzing X-rays, which can address two drawbacksof high cost and limited availability of testing kits. They stated that their result was unbia-sed because data leaking was avoided. In less than 5 seconds, it can discover a COVID-19positive patient. This methodology may address two issues: the high cost of testing kits andtheir scarcity.

Other Datasets and DenseNet:Zhu et al. (2020) created a deep neural network architecture for feature selection that

consisted of 6 fully linked dense layers and 56 clinical variables as inputs. As activationfunctions, they chose rectified linear units. The output layer used the sigmoid functionbecause they employed binary classification to classify patients into the expired and survived

27


groups. They also employed five-fold cross-validation to improve the model’s performance.The model prediction performance was utilized to rank the 56 features from the neuralnetwork by importance after fitting using permutation importance methods. The model’serror was examined by shuffling single features and selecting the characteristics with thehighest significance by comparing the model accuracy of testing with different features. Toavoid overfitting for prediction using the top (5) clinical variables or risk stratification scores,the neural network architecture for classification was reduced to a simple model with twofully connected dense layers, rather than six, once the features were ranked and the optimalnumber of features was determined. In this research, a sigmoid function was employed toactivate the output layer.

Using CT Scan Datasets and AlexNet:Turkoglu (2021) built an expert system based on chest X-ray images dubbed the CO-

VIDetectionNet model, which has a 100% success rate for the detection of COVID-19 usingpre-trained deep features ensemble as well as feature selection. The entire procedure is or-ganized into three steps. First, a pre-trained AlexNet architecture is employed for featureextraction. Second, the most efficient features are chosen using the Relief Algorithm based onthe k-nearest neighbor algorithm. Finally, these selected characteristics are classified usingthe Support Vector Machine Method.

InstaCovNet-19 is a robust model proposed by Gupta et al. (2021) that can identifyCOVID-19 and pneumonia from infected persons without any human influence. This pro-posed model is based on several models, including ResNet-101, Xception, NASNet, Incep-tionV3, and MobileNetv2. These are all pre-trained models that utilize an integrated stacktechnique to produce the most accurate results.

Ahmed et al. (2021) proposed a DL strategy (with five convolutional layers) based onchest X-ray images that can diagnose COVID-19 and Viral Pneumonia faster and more re-liably. This proposed model contained five convolutional layers and used data augmentationtechniques such as width shift, rotation, zoom, height shift, and shear on the images. In ad-dition, the complete dataset was subjected to five-fold cross-validation to access the model.Thus, this model offers a complete testing solution. Jain et al. (2020) developed a two-stagestrategy model to detect the presence of COVID-19 while distinguishing COVID-19 inducedpneumonia from healthy, bacterial pneumonia, and viral pneumonia cases in one study. TheResNet50 network was utilized to differentiate between normal pneumonia, viral pneumonia,and bacterial pneumonia. The presence of COVID-19 was then detected using ResNet-101in positive viral-induced pneumonia cases.

4.4 Graph Neural Network in COVID-19

Khalifa et al. (2020) recognized pneumonia chest x-ray using generative adversarial networks(GAN), which solved the over-fitting problem and created additional images from the da-taset, according to research. The efficiency of the suggested model was reflected in GAN,with Resnet18 achieving the best accuracy. The proposed model is divided into three stagesand is based on two DL models. GANs (Generative Adversarial Networks) is a model thatconsists of two major networks. The generating network is the first, while the discriminatornetwork is the second. Finally, in terms of testing accuracy measurement, Resnet18 (99%) isthe best deep transfer model. Toutiaee et al. (2021) compare multiple time-series prediction

28


techniques incorporating auxiliary variables. Their Spatio-temporal graph neural networksapproach forecast the course of the pandemic. Saha et al. (2021) have proposed a Graph Iso-morphic Network (GIN) based model named GraphCovidNet to detect COVID-19 by usingCT-scans, and CXRs of the Covid affected patients. They follow a GIN-based architecture.

By merging clinical, radiological, and biochemical data, Lassau et al. (2021) sought topredict hospitalized patients’ outcomes. They used a DL model to assess CT scan imagesand a radiologist report that included a semiquantitative description of CT scans to examinethe increased amount of information offered by CT scans. Rahimzadeh et al. (2021) detailedthe automated process for recognizing COVID 19 in chest CT scan pictures. They createda new dataset with 15589 images of COVID-affected patients and 48260 CT scan images.To investigate the lung view and avoid false detections, the authors mentioned an imageprocessing algorithm.

Wang et al. (2021) developed a method for detecting coronavirus by changing the in-ception model with transfer learning. They collected CT images from 1065 patients withmajor and minor illnesses. They employed both internal and external validation methods.On the internal validation set, the authors had an accuracy of 89.5%, while on the exter-nal validation set, they had 79.3%. Xiaowei et al. proposed a method for early screening ofCOVID-19 patients using DL techniques. They utilize multiple CNN models to stratify CTimage datasets and estimate the probability of infection among COVID-19 patients. Theyused a total of 618 CT scans where their proposed model reached an accuracy of 86.7%.

Lokwani et al. (2020) proposed a prospective method using a neural network to diagnoseCOVID-19 patients (positive or negative) by chest CT scans. This transfer learning-based2D segmentation model using the U-Net architecture achieved a sensitivity of 0.964 and aspecificity of 0.884. Xception was used as the encoder for U-Net. Also, they determined a logicfor changing over slice-level predictions to scan-level, which decreased the false positives. Wecollected Fig. 16 that represents their work.

Inf-Net is a semi-supervised model for Covid-19 Lung Infection segmentation proposed byFan et al. (2020a). Chest CT images are fed into Convolution layers to extract high resolutionand low-level characteristics, and sick regions can be automatically recognized using thismode. After that, they developed a module for edge attention. Then, they generated a globalmap Sg aggregating these features using a Parallel partial decoder (PPD). These featureswere then combined and fed into multiple reverse attention (RA) modules. To forecast thelung infection locations, the final output of RA was fed into a Sigmoid activation function.Modules for reverse attention and edge attention were also utilized. They’ve also extendedSemi-Inf-Net to a multi-class lung infection labeling model to provide more details for futureCovid-19 and lung infection treatments. They fed both the infection segmentation resultsprovided by Inf-Net and the CT images into FCN8s (or Multi-class U-Net) to improvemulti-class infection labeling accuracy.

Wang et al. (2020e) presented a weakly supervised 3D deep CNN dubbed DeCoVNet toDetect COVID-19 in a study for COVID-19 classification and lesion localization. Using the2D UNet architecture, the first retrieved the 3D lung border of each chest CT volume forlung area segmentation. All of the chest CT data was then tested on the pre-trained UNetmodel to obtain the lung masks. DeCoVNet was built using medical ground truth labels suchas COVID-positive and COVID-negative. Finally, the DeCoVNet was trained to generatepredictions using the testing data. They classified the lung region using an unsupervised

29


learning model and eliminated the errors. Their proposed DeCoVNet activation regionsshowed a high recall, but they predicted a lot of false positives. The chest CT volume wasthen subjected to the 3D Connected component technique. To get the response map, theyestimated the variance. Finally, they got the Covid-19 lesion localization result on the CAMactivation region with the most overlap.

Wang et al. (2021) attempted to develop an AI program that can effectively analyzerepresentative CT images to screen Covid-19 using DL Method. They built this model bymodifying the typical Inception Network, which is based on the transfer learning neural ne-twork, and fine-tuning the modified Inception (M-Inception) Model with pre-trained weights.To begin, they pre-processed all of the input photographs they had gathered from varioussources. Second, ROI images were extracted using the Computer Vision Model, which wassubsequently trained to remove features using the modified Inception Model. Finally, theyused a fully linked layer to perform classification and prediction.

Zhang et al. (2020a) used DL-based software (uAI Intelligent Assistant Analysis Sys-tem) to aid in the diagnosis and quantification of COVID-19 Pneumonia. A modified 3Dconvolution neural network and a combined V-Net with bottleneck structures are used inthis AI program. This software can segment anatomical lung structures to detect infectionlocations and determine the proportion of infection in both lungs, with lobes and segmentsaltered using custom HU ranges. He et al. (2020) proposed a sample-efficient, accurate DLSelf-Trans approach to learn robust and unbiased feature representations, which integratescontrasting self-supervised learning with transfer learning to reduce the risk of overfittingwhen diagnosing COVID-19 from CT scans. They also created a publicly accessible datasetwith hundreds of COVID-19-positive CT scans.

Fan et al. (2020a) suggested Inf-Net, a dynamic model of infection segmentation thataids in the detection of infected regions in a Covid-19 patient’s lung CT. First, they gatheredcharacteristics from the upper levels of the DL model using a PPD. The combined featureof Inf-Net and Semi-Inf-Net features take important data and generate an overview forproducing outputs. Then, a semi-supervised, named Semi-Inf-Net model was developed andused to overcome the deficit of exceptionally labeled datasets. Finally, the authors put CTscans into two convolutional layers to get higher-resolution images, and they used an edgeattention module to improve the model. They utilized Inf-Net, Semi-Inf-Net, PPD, RA, andEA modules, as well as FCN8s, U-Net, and U-Net++. Xu et al. (2020) utilized numerousCNN models to classify CT image datasets and figure the infection probability of COVID-19. These discoveries may extraordinarily aid the early screening of patients with COVID-19. Amyar et al. (2020) created an autonomous multi-task DL classification segmentationalgorithm to screen COVID-19 and segment lesions using chest CT imaging to categorizeCOVID vs. Normal vs. Other Infections and segment COVID lesions.

4.5 Other Techniques Used in COVID-19

Wu et al. (2021) presented DeepGLEAM. This COVID-19 forecasting hybrid model beatexisting ML models, such as the vector auto-regressive (VAR) model and a pure DL mo-del, in one to four weeks forward prediction. In short-term projections one to three weeksahead, it outperformed the GLEAM model. They looked at approaches including bootstrap,quantile regression, posterior sampling, and approximate Bayesian inference, all routinely

30


employed to quantify uncertainty. DeepGLEAM outperformed the competition in terms ofaccuracy and confidence intervals. This article Wan et al. (2021) developed a visual inter-pretative framework for understanding DL algorithms, which was then utilized to diagnoseCOVID-19. Their approach generates a comprehensive interpretation of the DL model fromvarious perspectives, including training patterns, diagnostic efficiency, trained features, fe-ature extraction, hidden nodes, diagnostic prediction analysis, and so on. Abdani et al.(2020) proposed a conditional generative adversarial network to generate synthetic imagesto supplement the limited amount of data available in the case of COVID-19 classificationin another paper. They also offered two DL models: a lightweight architecture is proportio-nal to the amount of data available, and a heavyweight architecture is proportional to theamount of data available.

Pham et al. (2021) presented a framework for screening compound phenotype to deter-mine the application for Covid-19 patients using a DL-based DeepCE model. The DeepCEmodel learns characteristics between chemical substructure-gene and gene-gene relationshipsvia a multiheaded attention mechanism, similar to the neural model. They also suggest anew augmentation technique for extracting meaningful information from configurations thataren’t useful. On the other hand, they underscore the model’s importance by using it to re-purpose COVID-19 drugs. To ensure public health safety and prevent COVID-19, Konget al. (2021) proposed an edge processing-based mask identification structure (ECMask)that can guarantee real-time execution on low-power camera gadgets in transport. It wasbased on three main stages: video reconstruction, face discovery, and mask identification.The authors recommend that the presented model be employed in public transport to pro-vide real-time performance on a camera that has been mounted. The authors used their BusDrive Monitoring and public datasets to train the model heavily.

Saba et al. (2021) developed a potential model for measuring and the optimal lockdownapproach to alleviating COVID-19’s causalities in one study. Various countries are nowimplementing three types of lockdown (partial, crowd, and total). Therefore, the over-fittingmodel authors studied three countries from each kind of type of lockdown. Zhang et al.(2020b) utilized chest X-rays to screen for viral Pneumonia using CAAD, a machine-drivenand accurate technique that relies less on labeled viral pneumonia data. Furthermore, theyhave suggested a confidence prediction module that uses an anomaly probability as themodel confidence to calculate the dependability of model diagnosis. Finally, they wantedto improve the screening sensitivity by predicting anomaly detection failures. Cohen et al.(2020) created a dataset claimed to be the first public COVID-19 image data collection,encompassing hundreds of frontal view X-rays and clinical metadata (survival, ICU stay,incubation events, blood tests, location, and clinical comments).

Tan et al. (2020) proposed to develop and validate a Non-focus area prediction model inthe early stages of COVID-19 Pneumonia. They used the Auto-ML method to excavate thetexture features of the first chest CT image and assess the model’s value in terms of the degreeof Non-focus area damage and clinical classification. Callejon-Leblic et al. (2021) wanted tobuild a complete ML model to investigate the value of taste and smell abnormalities inconjunction with other COVID-19 symptoms. Therefore, they used a multi-center case-control in which RT-PCR tested patients were informed of the existence and severity oftheir symptoms using visual analog scales. Mele and Magazzino (2021) mentioned two uniqueways to understand the connection between economic growth, pollution emersion, and Indian

31


Fig. 9: Chart demonstrating the usage percentage of CNN variants for COVID-19 research

people’s deaths from COVID-19. First, they applied the time series approach and annualdata from 1980 to 2018 and used stationarity and Toda-Yamamoto causality tests. Theyalso confirm an ML causal relationship between COVID-19 fatalities, CO2, PM 2.5, andNO2 in 25 major cities. Sáez et al. (2021) investigated significant biases from data sourcevariability to uncover severity subgroups based on symptoms and comorbidities. They usedthe nCov2019 public dataset. However, this variability can reduce the representativeness ofthe training data for the target populations and increase the complexity of the over-fittingmodel.

A semi-novel strategy to automating the identification of Pneumonia with a sophisticatedmodel was proposed in a study. They addressed reverse-transcription polymerase chain reac-tion (RT-PCR) by revealing a low-cost, quick method based on a DL model. Ibrahim et al.(2021) created an automated system for classifying Pneumonia utilizing the AlexNet modelwith transfer learning technique. First, the photos of the patients were categorized as follows:COVID-19, bacterial Pneumonia, non-COVID-19 viral Pneumonia, and routine. Then, theydid two-way (binary) classification, three-way classification, and four-way classification withthese categories. While most of their findings were impressive, the most well-known four-way type results drew the most interest. Their four-way classification reached an accuracyof 94.00% incorrectly identifying the classes.

5. Critical Analysis on COVID-19 Research

5.1 Most common techniques acquired

This chapter discusses the most commonly used framework of ML and DL Model techniquesfor COVID-19-related problems. An in-depth evaluation was performed, and the purposesfor using these specific methodologies as well as their statistical data, are given in Fig. 10,11, 12 and 9.

Among all the papers we have reviewed, a significant percentage (60%) of paper thathas been used extensively is DL-based publications. From Fig. 10 we can also see thatalmost 35% of the total articles are ML-based publications, and 5% of papers are from othermathematical models.

32


We noticed that many ML papers were used for resolving COVID-19 issues out of all thedocuments included throughout this research study. We provided a survey of ML models’most used framework usage between them depending upon those ML publications. Lookingat Fig. 11, we can say that hybrid models (Deep LSTM, Convolutional LSTM, Bi-DirectionalLSTM, CNN-LSTM, CNN RNN, IRRCNN, BCNN), Decision Tree, SVM, RF and LR,XGBoost all these frameworks are primarily used in 8% to 11.1% of the comprehensivepaper analysis. Other than utilizing these frameworks, SGD, NN, NLP, SVR, MLP, KNN,and other methods are used for about 7.4% to 2.5% papers. These frameworks are usedextensively for classification and getting better and acceptable accuracy. Other ML methods,such as Ensemble Algorithm, Catboost, GNN, Gradient boosting, Grad-CAM, and SGD,were used in 1.2% of the total paper. They have been utilized for statistical processing ofCOVID-19 data and also forecasting and predicting COVID-19 time-series data.

Correspondingly, for the framework that has been used in DL methods, we conductedan implementation review of the most commonly used techniques. A lot of frameworks, suchas DNN, CNN, RNN, GAN, LSTM, and so on, are widely implemented, and the percentageof their usage is in Fig. 12. We found that CNN’s are the most commonly implemented DLmethods for overcoming COVID-19 based challenges. CNN is the crucial framework thathas higher accomplishment in the radiology, clinical imaging area. Based on our study offrameworks of CNN, one of the most diverse frameworks that have been used in most ofthe papers, we discovered that AlexNet, ResNet-34, ResNet50, MobileNet-V2, Inception V3,ResNet-18, U-Net, VGG-16, DenseNet, FC-DenseNet103, DenseNet121,

Tabela 3: Details of the Papers that accommodate Deep Le-arning Techniques

References Input TypeMethodTechniqueUsed

NumberofImagesData


CrossValida-tion

Sethi et al.(2020)

ChestX-Ray

7 Depthwiseseparableconvol-utionslayers

940images

Rotation,shifting,zooming,horizontalflipping

NA

Nayak et al.(2021)

ChestX-Ray

Xception,ResNet50,MobileNet,InceptionV3

6249images NA NA

Jain et al.(2021)

ChestX-Ray

InceptionV3,Xception,ResNet

6432images

Rotation,Zooming,Horizontalflipping

NA

Oh et al.(2020)

ChestX-Ray

FC-DenseNet103,ResNet-18

502images NA NA

33




NumberofImagesData


CrossValida-tion

Jain et al.(2020)

ChestX-Ray

ResNet50,ResNet101

1215images

Rotation,Gaussianblur

5 fold

Wang et al.(2020d)

CTImage

DenseNet121-FPN,COVID-19Net

5372patients NA NA

Bai et al.(2020a)

CTImage EfficientNet B4 1186

patients

Flipping,scaling,rotation,brightnessandcontras-tmanipu-lations,noise,blurring

NA

Punn et al.(2020)

Time-seriesData

SVR, PR,DNN, RNN,LSTM

932,605cases NA NA

Ahmedet al. (2021)

ChestX-Ray CNN 1,389

images

Rotation,shear,zoom,widthshift,heightshift

5 fold

Pham(2021)

ChestX-Ray

AlexNet,GoogleNetandSqueezeNet

3666images NA NA

Xu and Meng CTImage

ModifiedInceptiontransfer-learningmodel

1,065images NA NA

34




NumberofImagesData


CrossValida-tion

Hall et al. (2020) ChestX-Ray

TransferlearmingwithVGG16andResnet50

455images

Horizontalflipping 10 fold

Abbasimehrand Paki(2021)

TimeSeriesData

ATT_BO,CNN_BO,LSTM_BO

Notdefined NA NA

Panwaret al.(2020a)

ChestX-ray,CT-Scan

TransferlearningwithVGG-19,Grad-CAM

5856images NA NA

Gupta et al.(2021)

ChestX-Ray

ResNet101,Xception,NASNet,InceptionV3,MobileNetv2withan Integratedstacktechnique

1088images

Shearing,zooming NA

Qjidaaet al. (2020)

ChestRadio-graphicImage

VGG16 300images

Randomflipping,scaling,shearing

NA

Sahlol et al.(2020)

ChestX-Ray FO-MPA 3435

images NA NA

Hussainet al. (2021)

CTScan,ChestX-ray

22-layerCNNarchitecture


35




NumberofImagesData


CrossValida-tion

Gianchandaniet al. (2020)

ChestX-ray

Ensemble modelusing VGG16and Dense-Net

—

Horizontalflipping,verticalflipping,rotationand sheertransfor-mation

NA

Kamal et al.(2021)

ChestX-ray

VGG19,InceptionV3,ResNet50,ResNet50V2,MobileNet,MobileNetV2,DenseNet121,NasNetMobile

760images NA NA

Zhang et al.(2020c)

ChestX-ray

Transferlearning,ResNet34

189images

Randomverticalflip,resizecrop,randomhorizontalflip androtation

NA

Turkoglu(2021)

ChestX-ray

AlexNet,KNN, SVM


Bai et al.(2020b)

CTImage LSTM, MLP 133

patients NA 5 fold

Heidariet al. (2021)

CXRImage VGG16 8,474

images

Shearing,zoom,rotation,width andheight shift,andhorizontalflip

NA

36




NumberofImagesData


CrossValida-tion

Alakus andTurkoglu(2020)

Textdata

ANN,CNN,LSTM,RNN,CNN-LSTM,CNN-RNN


Jamshidiet al. (2020)

RT-PCR

DeepLSTM,Convolu-tionalLSTM,Bi-direc-tionalLSTM

NA NA NA

Ghoshaland Tucker(2020)

ChestX-ray

BCNN,ResNet50V2

6009images

Rotation,ZCAwhitened,scaling,shifting,shearing,flippedhorizontallyandvertically

NA

Shastriet al. (2020)

Timeseriesdata

StackedLSTM,Bi-directioralLSTM,Convolu-tionalLSTM,MAPE

NA NA NA

Salmanet al. (2020)

X-rayimages

CNN, VGG16,Inceptionv3,ResNet50

260images NA NA

37




NumberofImagesData


CrossValida-tion

Zhu et al.(2020)

Textdata

Neuralnetworkpredictivemodel


Elsheikhet al. (2021)

Timeseriesdata

LSTM,comparedwithARIMA,NARANN

9048cases NA NA

Karakanisand Le-ontidis(2021)

ChestX-ray

cGAN,ResNet8,CNN

1255images

misa-lignedandslightlyrotated

NA

Fan et al.(2020a)

CTimages

Inf-Net,Semi-Inf-Net,RA,FCN8s,U-Net

1700images Resizing NA

Wang et al.(2020c)

ChestX-ray

ResNet-50,ResNet-101,ResNet-152

1853images NA NA

Abdaniet al. (2020)

ChestX-ray

MobileNet V1,MobileNet V2,MobileNet V3,ShuffleNet-V1,SquezeNet,DarkCOVID-Net,SPP-COVID-Net


Fan et al.(2020b)

ChestX-ray

AlexNet,MobileNetv2,ShuffleNet,SqueezeNet,Xception


38




NumberofImagesData


CrossValida-tion

Sun et al.(2020)

CTimages

VB-Net, GGO,XGboost,AFS-DF,LR, SVM,RF, NN,ElasticNet.


Shalbafet al. (2021)

CTscans

EfficientNets(B0-B5),NasNetLarge,NasNetMobile,InceptionV3,ResNet-50,SeResnet50,Xception,DenseNet121,ResNeXt50,Inception-ResNetV2

746images

Horizontallyandverticallyshift,randomrotation,fliphorizontally

NA

Rahmanet al. (2020)

CCTVimage

Imageprocessingand CNN

858images Resizing NA

Nath et al.(2020)

ChestCT, X-rayimages

CNN, GGO,Transferlearning

4561images

Resizing,Max pooling NA

Zeng et al.(2021) Text data

NLP,XGBoost,MLP,Bi-LSTM,SVR,

301,363patients NA NA

Sandu andKarim(2020)

CT scans

FastCapsNet,CapsNet,ResNet50,AlexNet

3000images NA NA

Das et al.(2021)

X-Rayimages

DenseNet201,ResNet50V2,Inceptionv3

1006images

Normalizing,resizing NA

39




NumberofImagesData


CrossValida-tion

Jain et al.(2021)

X-Rayimages

XCeption,InceptionNet3,ResNet,CNN

6432images Scaling NA

Perumalet al. (2021)

CT scan,CXR images

Resnet50,VGG16,InceptionV3

5,232images

Maxpooling 2d NA

Fan et al.(2020b)

X-rayimages

AlexNet,MobileNetv2,ShuffleNet,SqueezeNet,Xception


Narin et al.(2020)

X-rayimages

InceptionV3,ResNet50,ResNet101,ResNet152,Inception-ResNetV2


Punia et al.(2020) images ResNet-34,

ResNet-501122images

Rotating,zooming,heightshifting

NA

Gozes et al.(2020)

CT Scan-Images

Resnet-50,Deep CNN

6356Images NA NA

Bruneseet al. (2020)

ChestX-rays

VGG-16,Grad-CAM

6,523images

AveragePooling 2D,Flatten,Dense,Dropout,Dense

NA

Li et al.(2020)

CXRimages

FC-classifier,MobileNetV2,DenseNet-121,SqueezeNet

537images NA NA

Mason(2020) NA

It was aboutpatho-biology ofSARS-CoV-2virus

NA NA NA

40




NumberofImagesData


CrossValida-tion

Lokwaniet al. (2020)

CTscans

Xception,U-Net

275images NA NA

Zhang et al.(2020a)

RT-PCRresults,CT scans

It is asoftware

4675patients NA NA

Barbosa Jret al. (2020)

CXRimages

(1) SectraPACS(2) ITK-Snap

86patients

ADsegmentation NA

Ghoshaland Tucker(2020)

X-rayimages ResNet50V2 6009

images

Whitened,rotated,flippedhorizontally,vertically,scaled,shifted,sheared

NA

Khalifaet al. (2020)

X-rayimages

AlexNet,GoogleNet,SqueezNet,ResNet18

5863images NA NA

Wang et al.(2020a)

CXRimages

VGG-19,ResNet-50

13,975images NA NA

Hemdanet al. (2020)

X-rayimages

VGG19,DenseNet,InceptionV3,ResNetV2,Xception,MobileNetV2,Inception-ResNet-V2

50images NA NA

Hamidet al. (2020)

Texts,tweets GNN 6460

Separatingcomma,otherpunctuations

NA

41




NumberofImagesData


CrossValida-tion

Tu et al.(2020)

Texts,articles

ScispaCyNERmodel,BIONLP13CGcorpus

60,000papers

Semanticvisualizationtechniques

NA

He et al.(2020) CT scans

VGG-16 ,ResNet-18 ,ResNet-50,DenseNet-121,DenseNet-169,EfficientNetb0,EfficientNet-b1,CRNet

746images NA NA

Panwaret al.(2020b)

X-rayimages

VGG16Model

6388images

Rotation,flippedhorizontallyandvertically

NA

Alom et al.(2020)

CT image,X-ray IRRCNN 5216

images

Classspecificdataaugmentation

NA

Song et al.(2021) Images

ResNet-50,DenseNet,VGG16

1990images

Translationalandrotational

NA

Waheedet al. (2020)

CXRimages VGG16 1124

imagesSyntheticaugmentation NA

Yoo et al.(2020)

CXRimages ResNet18 1884

images

Horizontalflip,rotations,shifts

NA

GoogleNet, ImageNet, InceptionV3, Squeeze Net, and other models were used. FromFig. 9, ResNet algorithm used mostly in maximum papers. VGG is the second most usedframework, and DenseNet, AlexNet, XceptionNet CNN frameworks are most often employedfor COVID-19 based disputes. Inception, SqueezeNet, and MobileNet are also used in theimplementation of different CNN methods. So to do automated detection of Covid-19 andget an efficient accuracy, CNN frameworks are widely used.

42


Fig. 10: Percentage of paper published in machine learning, deep learning and other AIdomain

CNN is used widely in a lot of papers, almost 31%. Moreover, we can see from Fig.12 that LSTM was also primarily used in papers which are known as one of the mostefficient approaches to DL algorithms for time series analysis of different data. Besidesusing all these methods, authors have used Hybrid models, DNN, RNN, GAN, and othermathematical techniques. Based on our study of frameworks of CNN, one of the most diverseframeworks that has been used in most of the papers, we discovered that AlexNet, ResNet-34, ResNet50, MobileNet-V2, Inception V3, ResNet-18, U-Net, VGG-16, DenseNet, FC-DenseNet103, DenseNet121, GoogleNet, ImageNet, InceptionV3, Squeeze Net, and othermodels were used. From Fig. 9, ResNet algorithm used mostly in From Fig. 9, ResNetalgorithm used mostly in maximum papers. Other than that VGG is the second most usedframework, and DenseNet, AlexNet, XceptionNet CNN frameworks are most often employedfor COVID-19 based disputes. Inception, SqueezeNet, and MobileNet are also used in theimplementation of different CNN methods. So to do automated detection of Covid-19, andto get an efficient accuracy CNN frameworks are widely used.

5.2 Commonly used Data

With the new era of advancement, there is now tremendous change when we are getting theauthentic source of data, and now the collection of the dataset is quite large. This continuousgrowth of medical images, it is quite difficult to analyze by a medical expert, may have comemore variations from different expert and error-prone.

For automating the diagnosis process, the alternative solution is using ML and DL whe-re these techniques have played an important role in medical image segmentation, imageclassification, and computer-aided diagnosis. Based on this scenario, we have calculated thepercentage of a total of 100 papers. After estimating the percentage, it is noticed that imagedata is used in 63.6% of papers, text data is used in 17% of papers, time-series data is usedin 10.2% of papers, EHR data used in 5.7% of paper, and other data forms are used in 3.4%of papers. The results of this overall study are shown in Fig. 13. Here, Fig. 14 depicts thatthere is various input image data as well as others.

43


Fig. 11: The percentage ratio of different types of Machine learning methods in COVID-19research community

Fig. 12: The percentage ratio of different types of Deep learning methods in COVID-19research community

5.3 Machine Learning in COVID-19 Prediction and Detection

According to the findings, computational ML, and DL algorithm techniques have been ef-fectively used in the classification of clinical images through factors including the extractionof advanced features from spatial medical datasets and the capacity to differentiate betweenSars-Cov-2 and viral pneumonia. But the situation of pandemic’s ultimate uncertainty willmislead the goal of detection and prediction accuracy.

• The exact amount of people affected died or still tested COVID-19 positive cannot bepredicted with a 100% accuracy because of the false-positive and false-negative highratings and less COVID-19 detection test.

44


Fig. 13: The percentage of published COVID-19 papers with respect to the types of dataused in their research

Fig. 14: Usage of types of images data that are acquired by our the paper demonstrated inour survey

• Proper annotation of images are also affecting the COVID-19 prediction using ML andDL. As most of the detection and prediction is done by medical images, annotation ofthe image dataset should be properly done.

• The data that is currently available today is focused on COVID-19 tests that areconducted differently in various countries. They may not have to represent the wholeviral infection in detail, which may be the ML and DL model’s conceptual input. Toaccurately diagnose and detect this disease, and to do proper forecasting using MLand DL methods we need to implement accurate data.

5.4 Impact of Data Augmentation

We can use data augmentation to get more variant data, prevent overfitting for betterprediction, and perfectly train the model. By using data augmentation, it helps to increasethe diversity of data as well as help to avoid overfitting. It can be more helpful for smalldatasets by increasing the size of data for better image segmentation, classification, andobject detection. Before applying data augmentation, we need to understand the problemdomain. It can be applied as a pre-processing step before train the model after estimating

45


Fig. 15: Different types of Data augmentation techniques applied by the research papersillustrated in our survey

the percentage from Fig. 15, we can observe that powerful data augmentation techniques areused in many articles. Here, Rotation, Flipping, and Scaling have mainly used techniques.Also, there are having different data augmentation techniques along with others.

For developing a high-performance model, data augmentation can be used as a powerfultool as well as need rigorous problem analysis and domain knowledge.

6. Research Challenges

The following are the obstacles and issues related to DL implementations to monitor theCOVID-19 pandemic:

• Class-imbalance Problem: As Covid-19 pandemic outbreaks in 2020, “data collec-tion” was an obvious challenge for the authors. They find very little data on patientswith the Coronavirus. On the other hand, vast data of pneumonia patients or patientswith other lung diseases and healthy patients cause a class imbalance problem. Todeal with this issue, Zhang et al. (2020b) suggested an anomaly detector instead of aclassifier as the classifiers yield poor sensitivity performance for the class imbalanceproblem.

• Limited Data and Limitation Labeled Data: Most of the papers published in2020 faced this problem of limited data of COVID 19. They were trying to collectmore data not only with quality but also with labeling and annotation.

• Lack of Data Quality: Some of them collected data from online stores (books orpdf or newspapers) or from some random collection which impacts the quality of theresearch. Data should be collected from authentic sources like hospitals, clinics, or otherresearch papers. Good quality data can make the analysis more structured. Waheedet al. (2020) faced a challenge to create high-resolution samples from highly variabledata sets. Panwar et al. (2020b) have used only Posterior Anterior as a dataset.

46


• Selection Bias: During collecting data, considering 4g different variables is a must toenhance the quality of the data. Such as Cohen et al. (2020) did not consider immu-nosuppression status or ethnic background. For the primary research, it is a challengeto contemplate so many variables. Further research that is continuous analysis willimprove this condition. Selection bias is also seen while gathering publicly availableimages.

• Less Number of Features: Some of the papers demand that their work would bemore efficient if the number of features were increased. They will cope up with thischallenge in the further experiments they require.

• Single-center Design: Single-center data impacts hugely in the testing. The resultmay vary in the large dataset or a multi-centered dataset. Due to the single-centerdesign, their dataset is also smaller in size. Barbosa Jr et al. (2020) and Zhang et al.(2020a) found it difficult to collect multi-centered data. They felt that there was aneed to do in-depth research to make this more potential.

• Dataset on X-ray, not CT image: Comparing with computed tomography (CT),chest X-ray cannot come up with 3D anatomy and is less accurate than a PCR dia-gnostic. Chest Computed Tomography (CT) is a captivating option for patient mana-gement due to its spectacular portability, low cost, availability, and scalability. It playsan essential role in identifying earlier lung infections and monitoring disease develop-ment. Because of the low variance of the infection locales in CT images and enormousvarieties in the shapes and the places of lesions in various patients, the depiction ofcontamination regions in chest CT scans is sometimes also complicated.

• Unbalanced Data: For improved training in DL models and to obtain better featu-res, we need a vast amount of balanced dataset. Due to this problem, the authorsencountered many hurdles and had to apply different strategies to make it balanced,which is time-consuming.

• Hardware Limitations: Extensive data need for training intense models in less amo-unt of time. Accordingly, we need more powerful GPUs and high computing power.

• Lack of Diverse Datasets: Due to the lack of a diverse dataset, it is hard to analyzeand perfectly learn the models. When the model is fed with various training data, itprovides more discriminative information, making the model unique.

• Security of Privacy: By preserving privacy and maintaining the security of patient’smedical information, it is hard to collect a large number of medical images, whichhinders the training and validation of models.

• Time-intensive and Costly: High-quality data preparation is needed for the fast de-tection of COVID-19 when there are hidden biases in the dataset. Typically it requirestime-intensive, expensive human labor.

47


7. Future Research Directions

For fighting against the COVID-19, the new revolution of technology can be the savior ofour lives. Future COVID-19 research directions will guide healthcare planning, telemedicine,management strategies, and robust disease prediction at the earliest stage. In this regard,AI-based ML and DL systems contribute to the available resources where Transfer Learningplays a significant role with the help of unbalanced and limited datasets. In the medicalsector, DL can also develop a new robust architecture with the integration of optimizationalgorithms.

• Robust Development Architecture for COVID-19 and other Diseases: Anautomated, robust architecture can be built with the help of image classification usingX-ray and CT images for the detection of COVID-19 as well as other diseases, whichwill provide an end-to-end solution for fast detection.

• Necessity of Combined Data Repository: For further research development work,we need more data to combat the COVID-19 situation. Besides, this Combined DataRepository will help us to fight back with similar pandemics.

• Implementation of Medical Assistive in Real-World: For supporting the healthprofessionals, it is mandatory to implement this medical-assisting tool in hospitals,radiology clinics.

• Intelligent Robot: For assisting the COVID-19 infected patients without humanlabor, these robots are supposed to be used in hospitals to curb the spread of disease.

• Coronavirus detection by checking breathing sound: This can be a new andeasier area for detecting coronavirus. Scientists can focus on this for fast detection.

• AI in Biological Research: For biological research, this AI system can be used topredict the structure of proteins and genetic sequence.

It can be stated that more future direction and research are required for efficient identifica-tion and diagnosis.

Appendix A. Images

Remainder omitted in this sample. See http://www.jmlr.org/papers/ for full paper.

Literatura

Covid-19 vaccination tracker: Daily rates, statistics & updates, July 2021. URL https://www.pharmaceutical-technology.com/covid-19-vaccination-tracker/.

Coronavirus cases:, July 2021. URL https://www.worldometers.info/coronavirus/.

Hossein Abbasimehr and Reza Paki. Prediction of covid-19 confirmed cases combining deeplearning methods and bayesian optimization. Chaos, Solitons & Fractals, 142:110511,2021.

48

https://www.pharmaceutical-technology.com/covid-19-vaccination-tracker/

https://www.pharmaceutical-technology.com/covid-19-vaccination-tracker/

https://www.worldometers.info/coronavirus/


Fig. 16: (Left) original image and (right) corresponding predicted mask according to thismethod

Fig. 17: COVID-Net architecture

Siti Raihanah Abdani, Mohd Asyraf Zulkifley, and Nuraisyah Hani Zulkifley. A lightwe-ight deep learning model for covid-19 detection. In 2020 IEEE Symposium on IndustrialElectronics & Applications (ISIEA), pages 1–5. IEEE, 2020.

Parnian Afshar, Shahin Heidarian, Farnoosh Naderkhani, Anastasia Oikonomou, Konstan-tinos N Plataniotis, and Arash Mohammadi. Covid-caps: A capsule network-based frame-work for identification of covid-19 cases from x-ray images. Pattern Recognition Letters,138:638–643, 2020.

Faizan Ahmed, Syed Ahmad Chan Bukhari, and Fazel Keshtkar. A deep learning approachfor covid-19 8 viral pneumonia screening with x-ray images. Digital Government: Researchand Practice, 2(2):1–12, 2021.

Mohammed AA Al-Qaness, Ahmed A Ewees, Hong Fan, Laith Abualigah, and MohamedAbd Elaziz. Marine predators algorithm for forecasting confirmed cases of covid-19 initaly, usa, iran and korea. International journal of environmental research and publichealth, 17(10):3520, 2020.

Mohammed AA Al-Qaness, Amal I Saba, Ammar H Elsheikh, Mohamed Abd Elaziz, Re-hab Ali Ibrahim, Songfeng Lu, Ahmed Abdelmonem Hemedan, S Shanmugan, and Ah-

49


med A Ewees. Efficient artificial intelligence forecasting models for covid-19 outbreak inrussia and brazil. Process Safety and Environmental Protection, 149:399–409, 2021.

Ahmed Ali Mohammed Al-Saffar, Hai Tao, and Mohammed Ahmed Talab. Review of deepconvolution neural network in image classification. In 2017 International Conference onRadar, Antenna, Microwave, Electronics, and Telecommunications (ICRAMET), pages26–31. IEEE, 2017.

Talha Burak Alakus and Ibrahim Turkoglu. Comparison of deep learning approaches topredict covid-19 infection. Chaos, Solitons & Fractals, 140:110120, 2020.

Md Zahangir Alom, M M Shaifur Rahman, Mst Shamima Nasrin, Tarek M. Taha, and Vijay-an K. Asari. Covid_mtnet: Covid-19 detection with multi-task deep learning approaches,2020.

Amine Amyar, Romain Modzelewski, Hua Li, and Su Ruan. Multi-task deep learning basedct imaging analysis for covid-19 pneumonia: Classification and segmentation. Computersin Biology and Medicine, 126:104037, 2020.

Dilip Kumar Bagal, Arati Rath, Abhishek Barua, and Dulu Patnaik. Estimating the para-meters of susceptible-infected-recovered model of covid-19 cases in india during lockdownperiods. Chaos, Solitons & Fractals, 140:110154, 2020.

Harrison X Bai, Robin Wang, Zeng Xiong, Ben Hsieh, Ken Chang, Kasey Halsey, ThiMy Linh Tran, Ji Whae Choi, Dong-Cui Wang, Lin-Bo Shi, et al. Artificial intelligenceaugmentation of radiologist performance in distinguishing covid-19 from pneumonia ofother origin at chest ct. Radiology, 296,sahlol2020covid(3):E156–E165, 2020a.

Xiang Bai, Cong Fang, Yu Zhou, Song Bai, Zaiyi Liu, Liming Xia, Qianlan Chen, Yong-chao Xu, Tian Xia, Shi Gong, et al. Predicting covid-19 malignant progression with aitechniques. 2020b.

Pedro Baqui, Ioana Bica, Valerio Marra, Ari Ercole, and Mihaela van Der Schaar. Ethnicand regional variations in hospital mortality from covid-19 in brazil: a cross-sectionalobservational study. The Lancet Global Health, 8(8):e1018–e1026, 2020.

Eduardo Mortani Barbosa Jr, Warren B Gefter, Rochelle Yang, Florin C Ghesu, Siqi Liu,Boris Mailhe, Awais Mansoor, Sasa Grbic, Sebastian Piat, Guillaume Chabin, et al. Auto-mated detection and quantification of covid-19 airspace disease on chest radiographs: A no-vel approach achieving radiologist-level performance using a cnn trained on digital recon-structed radiographs (drrs) from ct-based ground-truth. arXiv preprint arXiv:2008.06330,2020.

Gopalkrishna Barkur and Giridhar B Kamath Vibha. Sentiment analysis of nationwidelockdown due to covid 19 outbreak: Evidence from india. Asian journal of psychiatry, 51:102089, 2020.

Mucahid Barstugan, Umut Ozkaya, and Saban Ozturk. Coronavirus (covid-19) classificationusing ct images by machine learning methods. arXiv preprint arXiv:2003.09424, 2020.

50


Muhammad Farhan Bashir, Benjiang Ma, Bushra Komal, Muhammad Adnan Bashir, Du-ojiao Tan, Madiha Bashir, et al. Correlation between climate indicators and covid-19pandemic in new york, usa. Science of the Total Environment, 728:138835, 2020.

Vaibhav Bhatnagar, Ramesh Chandra Poonia, Pankaj Nagar, Sandeep Kumar, VijanderSingh, Linesh Raja, and Pranav Dass. Descriptive analysis of covid-19 patients in thecontext of india. Journal of Interdisciplinary Mathematics, pages 1–16, 2020.

Pedro Henrique Borghi, Oleksandr Zakordonets, and João Paulo Teixeira. A covid-19 timeseries forecasting model based on mlp ann. Procedia Computer Science, 181:940–947, 2021.

Luca Brunese, Francesco Mercaldo, Alfonso Reginelli, and Antonella Santone. Explaina-ble deep learning for pulmonary disease and coronavirus covid-19 detection from x-rays.Computer Methods and Programs in Biomedicine, 196:105608, 2020.

María A Callejon-Leblic, Ramon Moreno-Luna, Alfonso Del Cuvillo, Isabel M Reyes-Tejero,Miguel A Garcia-Villaran, Marta Santos-Peña, Juan M Maza-Solano, Daniel I Martín-Jimenez, Jose M Palacios-Garcia, Carlos Fernandez-Velez, et al. Loss of smell and taste canaccurately predict covid-19 infection: A machine-learning approach. Journal of ClinicalMedicine, 10(4):570, 2021.

Zlatan Car, Sandi Baressi Šegota, Nikola Anđelić, Ivan Lorencin, and Vedran Mrzljak. Mo-deling the spread of covid-19 infection using a multilayer perceptron. Computational andmathematical methods in medicine, 2020, 2020.

Varun Chandola, Arindam Banerjee, and Vipin Kumar. Anomaly detection: A survey. ACMcomputing surveys (CSUR), 41(3):1–58, 2009.

Hui Juan Chen, Li Mao, Yang Chen, Li Yuan, Fei Wang, Xiuli Li, Qinlei Cai, Jie Qiu, JieTian, and Feng Chen. Machine learning-based ct radiomics model distinguishes covid-19from non-covid-19 pneumonia. 2021.

Yushi Chen, Xing Zhao, and Xiuping Jia. Spectral–spatial classification of hyperspectraldata based on deep belief network. IEEE Journal of Selected Topics in Applied EarthObservations and Remote Sensing, 8(6):2381–2392, 2015.

Muhammad EH Chowdhury, Tawsifur Rahman, Amith Khandakar, Rashid Mazhar, Mu-hammad Abdul Kadir, Zaid Bin Mahbub, Khandakar Reajul Islam, Muhammad SalmanKhan, Atif Iqbal, Nasser Al Emadi, et al. Can ai help in screening viral and covid-19pneumonia? IEEE Access, 8:132665–132676, 2020.

Joseph Paul Cohen, Paul Morrison, Lan Dao, Karsten Roth, Tim Q Duong, and MarzyehGhassemi. Covid-19 image data collection: Prospective predictions are the future. arXivpreprint arXiv:2006.11988, 2020.

Wesley Cota et al. Monitoring the number of covid-19 cases and deaths in brazil at municipaland federative units level. 2020.

Domenico Cucinotta and Maurizio Vanelli. Who declares covid-19 a pandemic. Acta BioMedica: Atenei Parmensis, 91(1):157, 2020.

51


Amit Kumar Das, Sayantani Ghosh, Samiruddin Thunder, Rohit Dutta, Sachin Agarwal,and Amlan Chakrabarti. Automatic covid-19 detection from x-ray images using ensemblelearning with convolutional neural network. Pattern Analysis and Applications, pages1–14, 2021.

Tiago de Melo and Carlos MS Figueiredo. A first public dataset from brazilian twitter andnews on covid-19 in portuguese. Data in brief, 32:106179, 2020.

Andreza Aruska de Souza Santos, Darlan da Silva Candido, William Marciel de Souza, LewisBuss, Sabrina L Li, Rafael HM Pereira, Chieh-Hsi Wu, Ester C Sabino, and Nuno RFaria. Dataset on sars-cov-2 non-pharmaceutical interventions in brazilian municipalities.Scientific data, 8(1):1–6, 2021.

Sinsuda Dechsupa, Suvichada Assawakosri, Suphattra Phakham, and Sittisak Honsawek.Positive impact of lockdown on covid-19 outbreak in thailand. Travel medicine and infec-tious disease, 36:101802, 2020.

Jayanthi Devaraj, Rajvikram Madurai Elavarasan, Rishi Pugazhendhi, GM Shafiullah, Su-mathi Ganesan, Ajay Kaarthic Jeysree, Irfan Ahmad Khan, and Eklas Hossain. Foreca-sting of covid-19 cases using deep learning models: Is it reliable and practically significant?Results in Physics, 21:103817, 2021.

Marzieh Edraki, Nazmul Karim, Nazanin Rahnavard, Ajmal Mian, and MubarakShah. Odyssey: Creation, analysis and detection of trojan models. arXiv preprintarXiv:2007.08142, 2020.

Mohamed Abd Elaziz, Khalid M Hosny, Ahmad Salah, Mohamed M Darwish, SongfengLu, and Ahmed T Sahlol. New machine learning method for image-based diagnosis ofcovid-19. Plos one, 15(6):e0235187, 2020.

Mohamed K Elhadad, Kin Fun Li, and Fayez Gebali. Detecting misleading information oncovid-19. Ieee Access, 8:165201–165215, 2020.

Ammar H Elsheikh, Amal I Saba, Mohamed Abd Elaziz, Songfeng Lu, S Shanmugan, T Mu-thuramalingam, Ravinder Kumar, Ahmed O Mosleh, FA Essa, and Taher A Shehabeldeen.Deep learning-based forecasting model for covid-19 outbreak in saudi arabia. Process Sa-fety and Environmental Protection, 149:223–233, 2021.

Deng-Ping Fan, Tao Zhou, Ge-Peng Ji, Yi Zhou, Geng Chen, Huazhu Fu, Jianbing Shen,and Ling Shao. Inf-net: Automatic covid-19 lung infection segmentation from ct images.IEEE Transactions on Medical Imaging, 39(8):2626–2637, 2020a.

Zeming Fan, Mudasir Jamil, Muhammad Tariq Sadiq, Xiwei Huang, and Xiaojun Yu. Explo-iting multiple optimizers with transfer learning techniques for the identification of covid-19patients. Journal of healthcare engineering, 2020, 2020b.

Muhammad Farooq and Abdul Hafeez. Covid-resnet: A deep learning framework for scre-ening of covid19 from radiographs. arXiv preprint arXiv:2003.14395, 2020.

52


Fernando Timoteo Fernandes, Tiago Almeida de Oliveira, Cristiane Esteves Teixeira, An-dre Filipe de Moraes Batista, Gabriel Dalla Costa, and Alexandre Dias Porto Chiavegat-to Filho. A multipurpose machine learning approach to predict covid-19 negative prognosisin são paulo, brazil. Scientific reports, 11(1):1–7, 2021.

Biraja Ghoshal and Allan Tucker. Estimating uncertainty and interpretability in deep lear-ning for coronavirus (covid-19) detection. arXiv preprint arXiv:2003.10769, 2020.

Neha Gianchandani, Aayush Jaiswal, Dilbag Singh, Vijay Kumar, and Manjit Kaur. Rapidcovid-19 diagnosis using ensemble deep transfer learning models from chest radiographicimages. Journal of ambient intelligence and humanized computing, pages 1–13, 2020.

Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep learning. MIT press, 2016.

Ophir Gozes, Maayan Frid-Adar, Hayit Greenspan, Patrick D Browning, Huangqi Zhang,Wenbin Ji, Adam Bernheim, and Eliot Siegel. Rapid ai development cycle for the corona-virus (covid-19) pandemic: Initial results for automated detection & patient monitoringusing deep learning ct image analysis. arXiv preprint arXiv:2003.05037, 2020.

Anunay Gupta, Shreyansh Gupta, Rahul Katarya, et al. Instacovnet-19: A deep learningclassification model for the detection of covid-19 patients using chest x-ray. Applied SoftComputing, 99:106859, 2021.

Rajan Gupta and Saibal Kumar Pal. Trend analysis and forecasting of covid-19 outbreakin india. MedRxiv, 2020.

Lawrence O Hall, Rahul Paul, Dmitry B Goldgof, and Gregory M Goldgof. Finding covid-19from chest x-rays using deep learning on a small dataset. arXiv preprint arXiv:2004.02060,2020.

Abdullah Hamid, Nasrullah Shiekh, Naina Said, Kashif Ahmad, Asma Gul, Laiq Hassan,and Ala Al-Fuqaha. Fake news detection in social media using graph neural networks andnlp techniques: A covid-19 use-case. arXiv preprint arXiv:2012.07517, 2020.

Arezoo Haratian, Hadi Fazelinia, Zeinab Maleki, Hao Wang, Mark Lewis, Russell Greiner,David Wishart, et al. Dataset of covid-19 outbreak and potential predictive features inthe usa. medRxiv, 2020.

Xuehai He, Xingyi Yang, Shanghang Zhang, Jinyu Zhao, Yichen Zhang, Eric Xing, andPengtao Xie. Sample-efficient deep learning for covid-19 diagnosis based on ct scans.MedRxiv, 2020.

Morteza Heidari, Seyedehnafiseh Mirniaharikandehei, Abolfazl Zargari Khuzani, GopichandhDanala, Yuchen Qiu, and Bin Zheng. Detecting covid-19 infected pneumonia from x-rayimages using a deep learning model with image preprocessing algorithm. In MedicalImaging 2021: Computer-Aided Diagnosis, volume 11597, page 115970V. InternationalSociety for Optics and Photonics, 2021.

53


Shahin Heidarian, Parnian Afshar, Nastaran Enshaei, Farnoosh Naderkhani, Moezedin JavadRafiee, Faranak Babaki Fard, Kaveh Samimi, S Farokh Atashzar, Anastasia Oikonomou,Konstantinos N Plataniotis, et al. Covid-fact: A fully-automated capsule network-basedframework for identification of covid-19 cases from chest ct scans. Frontiers in ArtificialIntelligence, 4, 2021.

Frank S Heldt, Marcela P Vizcaychipi, Sophie Peacock, Mattia Cinelli, Lachlan McLachlan,Fernando Andreotti, Stojan Jovanović, Robert Dürichen, Nadezda Lipunova, Robert AFletcher, et al. Early risk assessment for covid-19 patients from emergency departmentdata using machine learning. Scientific reports, 11(1):1–13, 2021.

Ezz El-Din Hemdan, Marwa A Shouman, and Mohamed Esmail Karar. Covidx-net: Aframework of deep learning classifiers to diagnose covid-19 in x-ray images. arXiv preprintarXiv:2003.11055, 2020.

Josh Holder. Tracking coronavirus vaccinations around the world, July 2021. URL https://www.nytimes.com/interactive/2021/world/covid-vaccinations-tracker.html.

Yeen Huang and Ning Zhao. Mental health burden for the public affected by the covid-19outbreak in china: Who will be the high-risk group? Psychology, health & medicine, 26(1):23–34, 2021.

Emtiaz Hussain, Mahmudul Hasan, Md Anisur Rahman, Ickjai Lee, Tasmi Tamanna, andMohammad Zavid Parvez. Corodet: A deep learning based classification for covid-19detection using chest x-ray images. Chaos, Solitons & Fractals, 142:110495, 2021.

Abdullahi Umar Ibrahim, Mehmet Ozsoz, Sertan Serte, Fadi Al-Turjman, and Polycarp Shi-zawaliyi Yakoi. Pneumonia classification using deep learning from chest x-ray imagesduring covid-19. Cognitive Computation, pages 1–13, 2021.

Aras M. Ismael and Abdulkadir Şengür. Deep learning approaches for covid-19 de-tection based on chest x-ray images. Expert Systems with Applications, 164:114054,2021. ISSN 0957-4174. doi: https://doi.org/10.1016/j.eswa.2020.114054. URL https://www.sciencedirect.com/science/article/pii/S0957417420308198.

Stefan Jaeger, Sema Candemir, Sameer Antani, Yì-Xiáng J Wáng, Pu-Xuan Lu, and GeorgeThoma. Two public chest x-ray datasets for computer-aided screening of pulmonarydiseases. Quantitative imaging in medicine and surgery, 4(6):475, 2014.

Govardhan Jain, Deepti Mittal, Daksh Thakur, and Madhup K Mittal. A deep learningapproach to detect covid-19 coronavirus with x-ray images. Biocybernetics and biomedicalengineering, 40(4):1391–1405, 2020.

Rachna Jain, Meenu Gupta, Soham Taneja, and D Jude Hemanth. Deep learning baseddetection and analysis of covid-19 on chest x-ray images. Applied Intelligence, 51(3):1690–1700, 2021.

Mohammad Jamshidi, Ali Lalbakhsh, Jakub Talla, Zdeněk Peroutka, Farimah Hadjilooei,Pedram Lalbakhsh, Morteza Jamshidi, Luigi La Spada, Mirhamed Mirmozafari, Mojgan

54

https://www.nytimes.com/interactive/2021/world/covid-vaccinations-tracker.html

https://www.nytimes.com/interactive/2021/world/covid-vaccinations-tracker.html

https://www.sciencedirect.com/science/article/pii/S0957417420308198

https://www.sciencedirect.com/science/article/pii/S0957417420308198


Dehghani, et al. Artificial intelligence and covid-19: deep learning approaches for diagnosisand treatment. IEEE Access, 8:109581–109595, 2020.

Stephen C Johnson. Hierarchical clustering schemes. Psychometrika, 32(3):241–254, 1967.

KC Kamal, Zhendong Yin, Mingyang Wu, and Zhilu Wu. Evaluation of deep learning-basedapproaches for covid-19 classification based on chest x-ray images. Signal, Image andVideo Processing, pages 1–8, 2021.

Stefanos Karakanis and Georgios Leontidis. Lightweight deep learning models for detectingcovid-19 from chest x-ray images. Computers in biology and medicine, 130:104181, 2021.

Sara Hosseinzadeh Kassani, Peyman Hosseinzadeh Kassasni, Michal J Wesolowski, Kevin ASchneider, and Ralph Deters. Automatic detection of coronavirus disease (covid-19) in x-ray and ct images: A machine learning-based approach. arXiv preprint arXiv:2004.10641,2020.

Nour Eldeen M Khalifa, Mohamed Hamed N Taha, Aboul Ella Hassanien, and Sally El-ghamrawy. Detection of coronavirus (covid-19) associated pneumonia based on generati-ve adversarial networks and a fine-tuned deep transfer learning model using chest x-raydataset. arXiv preprint arXiv:2004.01184, 2020.

Akib Mohi Ud Din Khanday, Syed Tanzeel Rabani, Qamar Rayees Khan, Nusrat Rouf, andMasarat Mohi Ud Din. Machine learning based approaches for detecting covid-19 usingclinical text data. International Journal of Information Technology, 12(3):731–739, 2020.

Xiangjie Kong, Kailai Wang, Shupeng Wang, Xiaojie Wang, Xin Jiang, Yi Guo, GuojiangShen, Xin Chen, and Qichao Ni. Real-time mask identification for covid-19: an edgecomputing-based deep learning framework. IEEE Internet of Things Journal, 2021.

Nathalie Lassau, Samy Ammari, Emilie Chouzenoux, Hugo Gortais, Paul Herent, MatthieuDevilder, Samer Soliman, Olivier Meyrignac, Marie-Pauline Talabard, Jean-Philippe La-marque, et al. Integrating deep learning ct-scan model, biological and clinical variables topredict severity of covid-19 patients. Nature communications, 12(1):1–11, 2021.

Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. nature, 521(7553):436–444, 2015.

Xin Li, Chengyin Li, and Dongxiao Zhu. Covid-mobilexpert: On-device covid-19 patienttriage and follow-up using chest x-rays. In 2020 IEEE International Conference on Bio-informatics and Biomedicine (BIBM), pages 1063–1067. IEEE, 2020.

Yen-Chin Liu, Rei-Lin Kuo, and Shin-Ru Shih. Covid-19: The first documented coronaviruspandemic in history. Biomedical journal, 43(4):328–333, 2020.

Mohamed Loey, Gunasekaran Manogaran, Mohamed Hamed N Taha, and Nour Eldeen MKhalifa. A hybrid deep transfer learning model with machine learning methods for facemask detection in the era of the covid-19 pandemic. Measurement, 167:108288, 2021.

55


Rohit Lokwani, Ashrika Gaikwad, Viraj Kulkarni, Aniruddha Pant, and Amit Kharat. Au-tomated detection of covid-19 from ct scans using convolutional neural networks. arXivpreprint arXiv:2006.13212, 2020.

Elisabeth Mahase. Covid-19: Novavax vaccine efficacy is 86% against uk variant and 60%against south african variant, 2021.

Efmi Maiyana et al. Pemanfaatan ann untuk prediksi penjualan online industri rumahanselama pandemi covid-19. Jurnal Sains dan Informatika: Research of Science and Infor-matic, 7(1):1–7, 2021.

Robert J Mason. Pathogenesis of covid-19 from a cell biology perspective, 2020.

Marco Mele and Cosimo Magazzino. Pollution, economic growth, and covid-19 deaths inindia: a machine learning evidence. Environmental Science and Pollution Research, 28(3):2669–2677, 2021.

Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Alex Graves, Ioannis Antonoglou, DaanWierstra, and Martin Riedmiller. Playing atari with deep reinforcement learning. arXivpreprint arXiv:1312.5602, 2013.

Ali Narin, Ceren Kaya, and Ziynet Pamuk. Automatic detection of coronavirus disease(covid-19) using x-ray images and deep convolutional neural networks. arXiv preprintarXiv:2003.10849, 2020.

Malaya Kumar Nath, Aniruddha Kanhe, and Madhusudhan Mishra. A novel deep learningapproach for classification of covid-19 images. In 2020 IEEE 5th International Conferenceon Computing Communication and Automation (ICCCA), pages 752–757. IEEE, 2020.

Soumya Ranjan Nayak, Deepak Ranjan Nayak, Utkarsh Sinha, Vaibhav Arora, and Ram Bi-las Pachori. Application of deep learning techniques for detection of covid-19 cases usingchest x-ray images: A comprehensive study. Biomedical Signal Processing and Control,64:102365, 2021.

Yujin Oh, Sangjoon Park, and Jong Chul Ye. Deep learning covid-19 features on cxr usinglimited training data sets. IEEE Transactions on Medical Imaging, 39(8):2688–2700, 2020.

Coşku Öksüz, Oğuzhan Urhan, and Mehmet Kemal Güllü. Ensemble-cvdnet: A deep lear-ning based end-to-end classification framework for covid-19 detection using ensembles ofnetworks. arXiv preprint arXiv:2012.09132, 2020.

Amir Hossein Panahi, Alireza Rafiei, and Alireza Rezaee. Fcod: Fast covid-19 detector basedon deep learning techniques. Informatics in medicine unlocked, 22:100506, 2021.

Harsh Panwar, PK Gupta, Mohammad Khubeb Siddiqui, Ruben Morales-Menendez, Pra-khar Bhardwaj, and Vaishnavi Singh. A deep learning and grad-cam based color visuali-zation approach for fast detection of covid-19 cases using chest x-ray and ct-scan images.Chaos, Solitons & Fractals, 140:110190, 2020a.

56


Harsh Panwar, PK Gupta, Mohammad Khubeb Siddiqui, Ruben Morales-Menendez, andVaishnavi Singh. Application of deep learning for fast detection of covid-19 in x-raysusing ncovnet. Chaos, Solitons & Fractals, 138:109944, 2020b.

Dhruv Patel, Vikram Kher, Bhushan Desai, Xiaomeng Lei, Steven Cen, Neha Nanda, AliGholamrezanezhad, Vinay Duddalwar, Bino Varghese, and Assad A Oberai. Machinelearning based predictors for covid-19 disease severity. Scientific Reports, 11(1):1–7, 2021.

Varalakshmi Perumal, Vasumathi Narayanan, and Sakthi Jaya Sundar Rajasekar. Detectionof covid-19 using cxr and ct images using transfer learning and haralick features. AppliedIntelligence, 51(1):341–358, 2021.

Thai-Hoang Pham, Yue Qiu, Jucheng Zeng, Lei Xie, and Ping Zhang. A deep learningframework for high-throughput mechanism-driven phenotype compound screening andits application to covid-19 drug repurposing. Nature Machine Intelligence, 3(3):247–257,2021.

Tuan D Pham. Classification of covid-19 chest x-rays with deep learning: new models orfine tuning? Health Information Science and Systems, 9(1):1–11, 2021.

Lan T Phan, Thuong V Nguyen, Quang C Luong, Thinh V Nguyen, Hieu T Nguyen, Hung QLe, Thuc T Nguyen, Thang M Cao, and Quang D Pham. Importation and human-to-human transmission of a novel coronavirus in vietnam. New England Journal of Medicine,382(9):872–874, 2020.

Sara Platto, Yanqing Wang, Jinfeng Zhou, and Ernesto Carafoli. History of the covid-19pandemic: origin, explosion, worldwide spreading. Biochemical and biophysical researchcommunications, 538:14–23, 2021.

Mohammad Pourhomayoun and Mahdi Shakibi. Predicting mortality risk in patients withcovid-19 using machine learning to help medical decision-making. Smart Health, 20:100178,2021.

Ravneet Punia, Lucky Kumar, Mohd Mujahid, and Rajesh Rohilla. Computer vision and ra-diology for covid-19 detection. In 2020 International Conference for Emerging Technology(INCET), pages 1–5. IEEE, 2020.

Narinder Singh Punn, Sanjay Kumar Sonbhadra, and Sonali Agarwal. Covid-19 epidemicanalysis using machine learning and deep learning algorithms. MedRxiv, 2020.

M Qjidaa, A Ben-Fares, Y Mechbal, H Amakdouf, M Maaroufi, B Alami, and H Qjidaa.Development of a clinical decision support system for the early detection of covid-19using deep learning based on chest radiographic images. In 2020 International Conferenceon Intelligent Systems and Computer Vision (ISCV), pages 1–6. IEEE, 2020.

Mohammad Rahimzadeh, Abolfazl Attar, and Seyed Mohammad Sakhaei. A fully automateddeep learning-based network for detecting covid-19 from a new and large lung ct scandataset. Biomedical Signal Processing and Control, page 102588, 2021.

57


Mohammad Marufur Rahman, Md Motaleb Hossen Manik, Md Milon Islam, Saifuddin Mah-mud, and Jong-Hoon Kim. An automated system to limit covid-19 using facial maskdetection in smart city network. In 2020 IEEE International IOT, Electronics and Me-chatronics Conference (IEMTRONICS), pages 1–5. IEEE, 2020.

Praveen Rai, Ballamoole Krishna Kumar, Vijaya Kumar Deekshit, Indrani Karunasagar,and Iddya Karunasagar. Detection technologies and recent developments in the diagnosisof covid-19 infection. Applied Microbiology and Biotechnology, pages 1–15, 2021.

Pranav Rajpurkar, Jeremy Irvin, Kaylie Zhu, Brandon Yang, Hershel Mehta, Tony Duan, Da-isy Ding, Aarti Bagul, Curtis Langlotz, Katie Shpanskaya, et al. Chexnet: Radiologist-levelpneumonia detection on chest x-rays with deep learning. arXiv preprint arXiv:1711.05225,2017.

Jawad Rasheed, Alaa Ali Hameed, Chawki Djeddi, Akhtar Jamil, and Fadi Al-Turjman.A machine learning-based framework for diagnosis of covid-19 from chest x-ray images.Interdisciplinary Sciences: Computational Life Sciences, 13(1):103–117, 2021.

Geetanjali Rathee, Sahil Garg, Georges Kaddoum, Yulei Wu, Dushantha Nalin K Jayakody,and Atif Alamri. Ann assisted-iot enabled covid-19 patient monitoring. IEEE Access, 9:42483–42492, 2021.

Matheus Henrique Dal Molin Ribeiro, Ramon Gomes da Silva, Viviana Cocco Mariani, andLeandro dos Santos Coelho. Short-term forecasting covid-19 cumulative confirmed cases:Perspectives for brazil. Chaos, Solitons & Fractals, 135:109853, 2020.

Salah Rifai, Pascal Vincent, Xavier Muller, Xavier Glorot, and Yoshua Bengio. Contractiveauto-encoders: Explicit invariance during feature extraction. In Icml, 2011.

Julien Riou and Christian L Althaus. Pattern of early human-to-human transmission of wu-han 2019 novel coronavirus (2019-ncov), december 2019 to january 2020. Eurosurveillance,25(4):2000058, 2020.

Santanu Roy, Gouri Sankar Bhunia, and Pravat Kumar Shit. Spatial prediction of covid-19epidemic using arima techniques in india. Modeling earth systems and environment, 7(2):1385–1391, 2021.

Tanzila Saba, Ibrahim Abunadi, Mirza Naveed Shahzad, and Amjad Rehman Khan. Machinelearning techniques to detect and forecast the daily total covid-19 infected and deaths casesunder different lockdown types. Microscopy Research and Technique, 2021.

Carlos Sáez, Nekane Romero, J Alberto Conejero, and Juan M García-Gómez. Potentiallimitations in covid-19 machine learning due to data source variability: A case study inthe ncov2019 dataset. Journal of the American Medical Informatics Association, 28(2):360–364, 2021.

Pritam Saha, Debadyuti Mukherjee, Pawan Kumar Singh, Ali Ahmadian, Massimiliano Fer-rara, and Ram Sarkar. Graphcovidnet: A graph neural network based model for detectingcovid-19 from ct scans and x-rays of chest. Scientific Reports, 11(1):1–16, 2021.

58


Ahmed T Sahlol, Dalia Yousri, Ahmed A Ewees, Mohammed AA Al-Qaness, Robertas Da-masevicius, and Mohamed Abd Elaziz. Covid-19 image classification using deep featuresand fractional-order marine predators algorithm. Scientific Reports, 10(1):1–15, 2020.

Hasim Sak, Andrew W Senior, and Françoise Beaufays. Long short-term memory recurrentneural network architectures for large scale acoustic modeling. 2014.

Sadman Sakib, Tahrat Tazrin, Mostafa M Fouda, Zubair Md Fadlullah, and Mohsen Guizani.Dl-crc: Deep learning-based chest radiograph classification for covid-19 detection: A novelapproach. IEEE Access, 8:171575–171589, 2020.

Ruslan Salakhutdinov and Geoffrey Hinton. Deep boltzmann machines. In Artificial intel-ligence and statistics, pages 448–455. PMLR, 2009.

Fatima M Salman, Samy S Abu-Naser, Eman Alajrami, Bassem S Abu-Nasser, and Belal AMAlashqar. Covid-19 detection using artificial intelligence. 2020.

Nitirajsingh Sandu and Shakir Karim. The application of fast capsnet computer vision indetecting covid-19. 2020.

Warren S Sarle. Neural networks and statistical models. 1994.

Rachna Sethi, Monica Mehrotra, and Dhaarna Sethi. Deep learning based diagnosis recom-mendation for covid-19 using chest x-rays images. In 2020 Second International Conferenceon Inventive Research in Computing Applications (ICIRCA), pages 1–4. IEEE, 2020.

Prabira Kumar Sethy, Santi Kumari Behera, Pradyumna Kumar Ratha, and Preesat Biswas.Detection of coronavirus disease (covid-19) based on deep features and support vectormachine. 2020.

Vruddhi Shah, Rinkal Keniya, Akanksha Shridharani, Manav Punjabi, Jainam Shah, andNinad Mehendale. Diagnosis of covid-19 using ct scan images and deep learning techniques.Emergency radiology, pages 1–9, 2021.

Ahmad Shalbaf, Majid Vafaeezadeh, et al. Automated detection of covid-19 using ensembleof transfer learning with deep convolutional neural network based on ct scans. Interna-tional journal of computer assisted radiology and surgery, 16(1):115–123, 2021.

Sourabh Shastri, Kuljeet Singh, Sachin Kumar, Paramjit Kour, and Vibhakar Mansotra.Time series forecasting of covid-19 using deep learning models: India-usa comparativecase study. Chaos, Solitons & Fractals, 140:110227, 2020.

Weiya Shi, Xueqing Peng, Tiefu Liu, Zenghui Cheng, Hongzhou Lu, Shuyi Yang, JiulongZhang, Mei Wang, Yaozong Gao, Yuxin Shi, et al. A deep learning-based quantitativecomputed tomography model for predicting the severity of covid-19: a retrospective studyof 196 patients. Annals of Translational Medicine, 9(3), 2021.

Yufang Shi, Ying Wang, Changshun Shao, Jianan Huang, Jianhe Gan, Xiaoping Huang,Enrico Bucci, Mauro Piacentini, Giuseppe Ippolito, and Gerry Melino. Covid-19 infection:the perspectives on immune responses, 2020.

59


Junji Shiraishi, Shigehiko Katsuragawa, Junpei Ikezoe, Tsuneo Matsumoto, Takeshi Kobay-ashi, Ken-ichi Komatsu, Mitate Matsui, Hiroshi Fujita, Yoshie Kodera, and Kunio Doi.Development of a digital image database for chest radiographs with and without a lungnodule: receiver operating characteristic analysis of radiologists’ detection of pulmonarynodules. American Journal of Roentgenology, 174(1):71–74, 2000.

Mohammad Shorfuzzaman and M Shamim Hossain. Metacovid: A siamese neural networkframework with contrastive loss for n-shot diagnosis of covid-19 patients. Pattern reco-gnition, 113:107700, 2021.

Lucas Silva and Dalson Figueiredo Filho. Using benford’s law to assess the quality of covid-19 register data in brazil. Journal of public health, 43(1):107–110, 2021.

Rajeev Kumar Singh, Rohan Pandey, and Rishie Nandhan Babu. Covidscreen: Explainabledeep learning framework for differential diagnosis of covid-19 using chest x-rays. NeuralComputing and Applications, pages 1–22, 2021.

Ying Song, Shuangjia Zheng, Liang Li, Xiang Zhang, Xiaodong Zhang, Ziwang Huang, Jian-wen Chen, Ruixuan Wang, Huiying Zhao, Yunfei Zha, et al. Deep learning enables accu-rate diagnosis of novel coronavirus (covid-19) with ct images. IEEE/ACM Transactionson Computational Biology and Bioinformatics, 2021.

A Spinelli and G Pellino. Covid-19 pandemic: perspectives on an unfolding crisis. Journalof British Surgery, 107(7):785–787, 2020.

R Sujath, Jyotir Moy Chatterjee, and Aboul Ella Hassanien. A machine learning foreca-sting model for covid-19 pandemic in india. Stochastic Environmental Research and RiskAssessment, 34:959–972, 2020.

Liang Sun, Zhanhao Mo, Fuhua Yan, Liming Xia, Fei Shan, Zhongxiang Ding, Bin Song,Wanchun Gao, Wei Shao, Feng Shi, et al. Adaptive feature selection guided deep forest forcovid-19 classification with chest ct. IEEE Journal of Biomedical and Health Informatics,24(10):2798–2805, 2020.

Richard S Sutton and Andrew G Barto. Reinforcement learning: An introduction. MITpress, 2018.

Kenji Suzuki. Overview of deep learning in medical imaging. Radiological physics andtechnology, 10(3):257–273, 2017.

Hui-Bin Tan, Fei Xiong, Yuan-Liang Jiang, Wen-Cai Huang, Ye Wang, Han-Han Li, TaoYou, Ting-Ting Fu, Ran Lu, and Bi-Wen Peng. The study of automatic machine learningbase on radiomics of non-focus area in the first chest ct of different clinical types of covid-19 pneumonia. Scientific reports, 10(1):1–10, 2020.

Mohammadhossein Toutiaee, Xiaochuan Li, Yogesh Chaudhari, Shophine Sivaraja, Aish-warya Venkataraj, Indrajeet Javeri, Yuan Ke, Ismailcem Arpinar, Nicole Lazar, andJohn Miller. Improving covid-19 forecasting using exogenous variables. arXiv preprintarXiv:2107.10397, 2021.

60


Jingxuan Tu, Marc Verhagen, Brent Cochran, and James Pustejovsky. Exploration anddiscovery of the covid-19 literature through semantic visualization. arXiv preprintarXiv:2007.01800, 2020.

Muammer Turkoglu. Covidetectionet: Covid-19 diagnosis system based on x-ray imagesusing features selected from pre-learned deep features ensemble. Applied Intelligence, 51(3):1213–1226, 2021.

Jijo Pulickiyil Ulahannan, Nikhil Narayanan, Nishad Thalhath, Prem Prabhakaran, Sre-ekanth Chaliyeduth, Sooraj P Suresh, Musfir Mohammed, E Rajeevan, Sindhu Joseph,Akhil Balakrishnan, et al. A citizen science initiative for open data and visualization ofcovid-19 outbreak in kerala, india. Journal of the American Medical Informatics Associa-tion, 27(12):1913–1920, 2020.

Elise van Nederveen Meerkerk. Segmentation in the pre-industrial labour market: women’swork in the dutch textile industry, 1581–1810. International review of social history, 51(2):189–216, 2006.

Abdul Waheed, Muskan Goyal, Deepak Gupta, Ashish Khanna, Fadi Al-Turjman, and Pláci-do Rogerio Pinheiro. Covidgan: data augmentation using auxiliary classifier gan for im-proved covid-19 detection. Ieee Access, 8:91916–91923, 2020.

Yuchai Wan, Hongen Zhou, and Xun Zhang. An interpretation architecture for deep learningmodels with the application of covid-19 diagnosis. Entropy, 23(2):204, 2021.

Linda Wang, Zhong Qiu Lin, and Alexander Wong. Covid-net: A tailored deep convolutionalneural network design for detection of covid-19 cases from chest x-ray images. ScientificReports, 10(1):1–12, 2020a.

Lucy Lu Wang, Kyle Lo, Yoganand Chandrasekhar, Russell Reas, Jiangjiang Yang, DarrinEide, Kathryn Funk, Rodney Kinney, Ziyang Liu, William Merrill, et al. Cord-19: Thecovid-19 open research dataset. ArXiv, 2020b.

Ningwei Wang, Hongzhe Liu, and Cheng Xu. Deep learning for the detection of covid-19using transfer learning and model integration. In 2020 IEEE 10th International Conferenceon Electronics Information and Emergency Communication (ICEIEC), pages 281–284.IEEE, 2020c.

Shuai Wang, Bo Kang, Jinlu Ma, Xianjun Zeng, Mingming Xiao, Jia Guo, Mengjiao Cai,Jingyi Yang, Yaodong Li, Xiangfei Meng, et al. A deep learning algorithm using ct imagesto screen for corona virus disease (covid-19). European Radiology, pages 1–9, 2021.

Shuo Wang, Yunfei Zha, Weimin Li, Qingxia Wu, Xiaohu Li, Meng Niu, Meiyun Wang,Xiaoming Qiu, Hongjun Li, He Yu, et al. A fully automatic deep learning system forcovid-19 diagnostic and prognostic analysis. European Respiratory Journal, 56(2), 2020d.

Xiaosong Wang, Yifan Peng, Le Lu, Zhiyong Lu, Mohammadhadi Bagheri, and Ronald MSummers. Chestx-ray8: Hospital-scale chest x-ray database and benchmarks on weakly-supervised classification and localization of common thorax diseases. In Proceedings ofthe IEEE conference on computer vision and pattern recognition, pages 2097–2106, 2017.

61


Xinggang Wang, Xianbo Deng, Qing Fu, Qiang Zhou, Jiapei Feng, Hui Ma, Wenyu Liu,and Chuansheng Zheng. A weakly-supervised framework for covid-19 classification andlesion localization from chest ct. IEEE transactions on medical imaging, 39(8):2615–2625,2020e.

Svante Wold, Kim Esbensen, and Paul Geladi. Principal component analysis. Chemometricsand intelligent laboratory systems, 2(1-3):37–52, 1987.

Salomón Wollenstein-Betech, Amanda AB Silva, Julia L Fleck, Christos G Cassandras, andIoannis Ch Paschalidis. Physiological and socioeconomic characteristics predict covid-19mortality and resource utilization in brazil. PloS one, 15(10):e0240346, 2020.

Dongxia Wu, Liyao Gao, Xinyue Xiong, Matteo Chinazzi, Alessandro Vespignani, Yi-AnMa, and Rose Yu. Deepgleam: A hybrid mechanistic and deep learning model for covid-19 forecasting, 2021.

Bo Xu and Xiangfei Meng. A deep learning algorithm using ct images to screen for coronavirus disease (covid-19). Preprint.

Xiaowei Xu, Xiangao Jiang, Chunlian Ma, Peng Du, Xukun Li, Shuangzhi Lv, Liang Yu,Qin Ni, Yanfei Chen, Junwei Su, et al. A deep learning system to screen novel coronavirusdisease 2019 pneumonia. Engineering, 6(10):1122–1129, 2020.

Ramjeet Singh Yadav. Data analysis of covid-2019 epidemic using machine learning methods:a case study of india. International Journal of Information Technology, 12:1321–1330,2020.

Yuki Yamada, Dominik-Borna Ćepulić, Tao Coll-Martín, Stéphane Debove, Guillaume Gau-treau, Hyemin Han, Jesper Rasmussen, Thao P Tran, Giovanni A Travaglino, and AndreasLieberoth. Covidistress global survey dataset on psychological and behavioural consequ-ences of the covid-19 outbreak. Scientific data, 8(1):1–23, 2021.

Xingyi Yang, Xuehai He, Jinyu Zhao, Yichen Zhang, Shanghang Zhang, and Pengtao Xie.Covid-ct-dataset: a ct image dataset about covid-19. arXiv preprint arXiv:2003.13865,2020.

Haochen Yao, Nan Zhang, Ruochi Zhang, Meiyu Duan, Tianqi Xie, Jiahui Pan, Ejun Peng,Juanjuan Huang, Yingli Zhang, Xiaoming Xu, et al. Severity detection for the coronavirusdisease 2019 (covid-19) patients using a machine learning model based on the blood andurine tests. Frontiers in cell and developmental biology, 8:683, 2020.

Seung Hoon Yoo, Hui Geng, Tin Lok Chiu, Siu Ki Yu, Dae Chul Cho, Jin Heo, Min SungChoi, Il Hyun Choi, Cong Cung Van, Nguen Viet Nhung, et al. Deep learning-baseddecision-tree classifier for covid-19 diagnosis from chest x-ray imaging. Frontiers in medi-cine, 7:427, 2020.

Vasilios Zarikas, Stavros G Poulopoulos, Zoe Gareiou, and Efthimios Zervas. Clusteringanalysis of countries using the covid-19 cases dataset. Data in brief, 31:105787, 2020.

62


Tahmina Zebin and Shahadate Rezvy. Covid-19 detection and disease progression visuali-zation: Deep learning on chest x-rays for classification and coarse localization. AppliedIntelligence, 51(2):1010–1021, 2021.

Wenhuan Zeng, Anupam Gautam, and Daniel H Huson. On the application of advancedmachine learning methods to analyze enhanced, multimodal data from persons infectedwith covid-19. Computation, 9(1):4, 2021.

Hai-tao Zhang, Jin-song Zhang, Hai-hua Zhang, Yan-dong Nan, Ying Zhao, En-qing Fu,Yong-hong Xie, Wei Liu, Wang-ping Li, Hong-jun Zhang, et al. Automated detectionand quantification of covid-19 pneumonia: Ct imaging analysis by a deep learning-basedsoftware. European journal of nuclear medicine and molecular imaging, 47(11):2525–2532,2020a.

Jianpeng Zhang, Yutong Xie, Guansong Pang, Zhibin Liao, Johan Verjans, Wenxing Li,Zongji Sun, Jian He, Yi Li, Chunhua Shen, et al. Viral pneumonia screening on chestx-rays using confidence-aware anomaly detection. IEEE transactions on medical imaging,2020b.

Ruochi Zhang, Zhehao Guo, Yue Sun, Qi Lu, Zijian Xu, Zhaomin Yao, Meiyu Duan, ShuaiLiu, Yanjiao Ren, Lan Huang, et al. Covid19xraynet: A two-step transfer learning modelfor the covid-19 detecting problem based on a limited number of chest x-ray images.Interdisciplinary Sciences: Computational Life Sciences, 12(4):555–565, 2020c.

Jie Zhao, Wei Cui, and Bao-ping Tian. The potential intermediate hosts for sars-cov-2.Frontiers in microbiology, 11, 2020.

Tao Zhou, Huiling Lu, Zaoli Yang, Shi Qiu, Bingqiang Huo, and Yali Dong. The ensembledeep learning model for novel covid-19 on ct images. Applied Soft Computing, 98:106885,2021.

Jocelyn S Zhu, Peilin Ge, Chunguo Jiang, Yong Zhang, Xiaoran Li, Zirun Zhao, LimingZhang, and Tim Q Duong. Deep-learning artificial intelligence analysis of clinical variablespredicts mortality in covid-19 patients. Journal of the American College of EmergencyPhysicians Open, 1(6):1364–1373, 2020.

Yazeed Zoabi, Shira Deri-Rozov, and Noam Shomron. Machine learning-based prediction ofcovid-19 diagnosis based on symptoms. npj digital medicine, 4(1):1–5, 2021.

63

arXiv:2111.09537v1 [cs.LG] 18 Nov 2021

Documents

Transcript of arXiv:2111.09537v1 [cs.LG] 18 Nov 2021