key: cord-0200326-dw4bprwr
authors: Sarosh, Parsa; Parah, Shabir A.; Mansur, Romany F; Bhat, G. M.
title: Artificial Intelligence for COVID-19 Detection -- A state-of-the-art review
date: 2020-11-25
journal: nan
DOI: nan
sha: eb9177c512b918cf552b63e8a7e5a2ff217e6a5a
doc_id: 200326
cord_uid: dw4bprwr

The emergence of COVID-19 has necessitated many efforts by the scientific community for its proper management. An urgent clinical reaction is required in the face of the unending devastation being caused by the pandemic. These efforts include technological innovations for improvement in screening, treatment, vaccine development, contact tracing and, survival prediction. The use of Deep Learning (DL) and Artificial Intelligence (AI) can be sought in all of the above-mentioned spheres. This paper aims to review the role of Deep Learning and Artificial intelligence in various aspects of the overall COVID-19 management and particularly for COVID-19 detection and classification. The DL models are developed to analyze clinical modalities like CT scans and X-Ray images of patients and predict their pathological condition. A DL model aims to detect the COVID-19 pneumonia, classify and distinguish between COVID-19, Community-Acquired Pneumonia (CAP), Viral and Bacterial pneumonia, and normal conditions. Furthermore, sophisticated models can be built to segment the affected area in the lungs and quantify the infection volume for a better understanding of the extent of damage. Many models have been developed either independently or with the help of pre-trained models like VGG19, ResNet50, and AlexNet leveraging the concept of transfer learning. Apart from model development, data preprocessing and augmentation are also performed to cope with the challenge of insufficient data samples often encountered in medical applications. It can be evaluated that DL and AI can be effectively implemented to withstand the challenges posed by the global emergency

The preliminary cases of COVID-19 pneumonia were identified in China (Wuhan District) in December 2019 [1, 2] . In January 2020, the World Health Organisation (WHO) declared COVID-19 as a public health emergency because of numerous cases being spread around the world [3] . As per a report of the WHO, more than 200 countries have been affected by the COVID-19 pandemic [4, 5] . Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-COV-2) virus that causes the COVID-19 infectious disease is a single-stranded RNA virus. The total number of cases as of 25 November 2020 GMT 13:03 is 60,108, 193 with deaths being equal to 1,414,925 [6] . Despite restrictions in all spheres of life including emphasis on cleanliness and social distancing, still the number of cases is increasing rapidly. The impact is more on In this paper, we present different AI-based models which help in the various aspects of the overall COVID-19 management. We further illustrate in detail the different models developed for Classification, detection, and delineation of the COVID-19 infection using Xray and CT images and the performance evaluation of the different models.

The developed models can be distinguished based on the modality utilized by the DLframework [26] . The DL models can be broadly classified as employing either the X-Ray image or the CT scan for COVID-19 diagnosis. Furthermore, the developed models are either custom made for COVID-19 detection or utilize pre-trained models as backbone as shown in fig 2. Some researchers have used off-the-shelf models pre-trained on large image databases like ImageNet dataset and repurposed them for COVID-19 detection and classification [27] . The DL system can be an off-the-shelf pre-trained model used directly like the AlexNet, Visual Geometry Group (VGG-16/19), U-Net, ResNet, Xception, and GoogLeNet [28] . The pretrained models can be further trained using smaller datasets often encountered in medical applications with the help of deep transfer learning. Other customized CNN models have also been evaluated which include the WOA-CNN, CRNet, DeCoVNet, COVID-Net, CoroNet among others [29] . The Xray images are simple to develop because of the availability of the prevalent necessary setup in almost all medical centers and are less expensive for the patient. However, the detection of COVID-19 from Xray images is difficult to perform because of subtle visual indicators. On the other hand, 3D-CT scans are developed using sophisticated machinery and are more expensive to generate, but they can provide a detailed 3D-pulmonary view. Furthermore, the sensitivity of CT images is much higher than that of RT-PCR testing [30] . According to a study conducted in Wuhan, the sensitivity of RT-PCR for the COVID-19 infection rate was 71% and CT images were about 98%. The high false-negative rate of RT-PCR tests indicate less efficiency in identifying the COVID-19 affected patient. This leads to delay in the administration of proper treatment to the patient and can also lead to more spreading of the virus because of a lack of proper identification. Therefore, it is concluded that the analysis of medical images like X-rays and CT scans can provide a more accurate alternative diagnostic tool for COVID-19 pneumonia [31, 32] . The clinical findings from CT and Xray images include bilateral and peripheral ground-glass opacities (GGO), consolidation and crazy-paving pattern. However, manual delineation of the affected areas in the lungs by radiologists can be a very hectic and timeconsuming process and particularly for the enormous number of patients arising from this pandemic [33] . The DL architectures can extract high-level features from both modalities and can be used to detect and diagnose COVID-19 pneumonia. The AI-based system can assist radiologists and help provide decision support to the clinicians in this pandemic for patient triage and treatment [34] .

According to the WHO guidelines, the spread of the virus can be contained through precautions like frequent cleansing of hands, use of personal protective equipment like gloves and masks, and Social distancing [35] . These protective measures can save others who come in contact with a person having the virus and can stop its further spread. The virus is mainly transmitted via discharges through the mouth and nose during coughing or sneezing [36] . Proper identification of the person who himself is infected and the ones he has interacted with can restrict the further spread of the SARS-Cov-2 virus. It is an important step in this regard that digital contact tracing should be implemented such that people who have come in contact with the virus can be identified and further spread of the virus can be contained [37] . Along with AI, DL, and ML, the use of various integrated technologies like IoT and communication systems like Bluetooth and GPS can be sought for this purpose [38, 39] . Many countries have created platforms for digital contact tracing which are illustrated in table 1. 

Forecasting models are developed to predict the growth of COVID-19 cases that might arise soon. This information will help in preparing the healthcare organizations and government institutions to generate and allocate resources like hospital beds, PPEs for healthcare workers, Ventilators, and Intensive care units to critical patients before the actual scenario arises [40, 41] . Vinay Kumar Reddy Chimmula et al. [42] present a DL-based approach to forecast the future COVID-19 cases in Canada using the datasets provided by John Hopkins University and Canadian health authorities. The forecasting method is based on a Long Short-term Memory (LSTM) network. Their transmission model predicted that the outbreak in Canada will have an ending point around June 2020. They conclude that the rate of transmission in Canada is taking a linear trend contrary to the exponential trend seen in the USA. Domenico Benvenuto et al. [43] propose an econometric model to predict the spread of the virus. They utilize the ARIMA model for prediction of epidemiological trend of occurrence of cases with the help of datasets provided by John Hopkins University. The ARIMA (1,0,4) was chosen to be the best model to calculate the prevalence of COVID-19 and ARIMA (1,0,3) has the best performance for calculating the incidence of COVID-19. Li Yan et al. [44] have proposed an XGBoost MLbased model that can select biomarkers that can predict the mortality of individuals almost ten days in advance with more than 90% accuracy. These biomarkers include lymphocyte, lactic dehydrogenase (LDH), and high-sensitivity C-reactive protein (hs-CRP). They evaluate that high values of LDH acts as an indicator for the identification of patients requiring immediate medical attention. Tanujit Chakraborty et al. [45] present a data-driven method to forecast the future cases that might arise. The data analysis has been done for various countries like France, Canada, India, South Korea, and the UK and also include risk assessment in terms of fatality rate. The method is based on a moving average model and wavelet-based forecasting model for making predictions (10 days prior) about the total number of confirmed cases. This will help prepare and allocate necessary services and equipment and can also be an early warning for the people, healthcare professionals, and government organizations. Matheus Henrique Dal Molin Ribeiro et al. [46] evaluate several models for time-series forecasting with 1-3 and 6 days ahead. The evaluated models include Autoregressive Integrated Moving Average (ARIMA), Random Forest (RF), ridge regression (RIDGE), Cubist regression (CUBIST), Support-Vector Regression (SVR), and stacking-ensemble learning. The best performance is offered by the SVR model and worst by the RF models in terms of evaluation parameters like improvement index, and mean absolute error among others.

Yet another area where AI and DL can help in assisting healthcare professionals is to identify the use of existing drugs for the treatment of COVID-19 or develop new vaccines [47] . Yi-Yu Ke et al. [48] have developed an AI-based method to repurpose the potential older drugs for COVID-19 treatment. The older drugs with anti-coronavirus behavior were identified by the use of an AI platform and two learning databases. One of the databases had compounds with properties against SARS-CoV, SARS-CoV2, Influenza virus, and the other database contained 3C-like protease inhibitors. The AI-assisted method can rapidly identify which older drugs can be repurposed for the treatment of COVID-19. They isolate 8 drugs including vismodegib, conivaptan, brequinar, bedaquiline, celecoxib, gemcitabine, clofazimine, tolcapone with potential antiviral properties. Sean Ekins et al. [49, 50] provide a detailed explanation of the drug discovery method for Zika and Ebola viruses and further illustrate that there might be lacunas in the drug discovery process. They further explain the attempts of drug discovery for COVID-19 and whether existing antiviral therapeutics can be repurposed for SARS-CoV-2 virus. Bo Ram Beck et al. [51] have presented a DL model called Molecule Transformer-Drug Target Interaction (MT-DTI) to identify existing drugs that can be used against the SARS-CoV-2 virus. They have identified a list of existing antiviral drugs that can be considered for the treatment of COVID-19. According to the MT-DTI model, the atazanavir is considered as the best compound against SARS-CoV-2 3C like protease. They also propose that antiviral agents like Kaletra could be used as a treatment for COVID-19.

The spread of COVID-19 can be limited by the use of Machine Learning (ML) and Deep Learning systems for early detection and screening of patients [52] . With the help of DL basedmodel, the COVID-19 pneumonia can be easily detected from CT, ultrasound, and Xray images [53, 54] . These systems will perform classification between COVID-19, normal, bacterial pneumonia and will also segment and quantify the infection sites [55] . In brief, the main tasks to be undertaken by a DL-based system are enlisted as follows:

1. Binary Classification to distinguish COVID-19 cases from healthy cases for proper detection and diagnosis. 2. Multi-class classification to correctly classify COVID-19 pneumonia, communityacquired pneumonia, bacterial pneumonia, and healthy cases. 3. Delineation of the affected regions in the lungs using the process of segmentation. 4. Quantification of the level of infection in the lungs using deep learning models. 5. Overall decision support for clinicians in the healthcare systems.

For the task of binary classification, the supervised DL models are provided with suitable datasets of normal and COVID-19 pneumonia images and class labels [56, 57] . These models extract deep features from the images and map them to their respective category's labels. The fully trained models are then used for the classification of new Xray or CT images that were not used in training. According to the survey conducted, most of the research works employ either of the two modalities namely the Xray images and CT scans for automatic detection using Deep Learning (DL) architectures [58, 59] .

The first step in the development of an automatic COVID-19 diagnosis system is the collection of data that is required to train the DL models. In this scenario, the CT scans and X-Ray images form the most important modalities that can be used. The collected data is then pre-processed and augmented by operations like normalization, contrast enhancement, rotation, horizontal and vertical flip so that the CNN model can extract relevant features. Joseph Paul Cohen et al. [60] 

The DL models perform end-to-end training by a sequence of operations like convolution and pooling on the input modality like CT or X-Ray image [62] . Furthermore, various types of activation functions like ReLU, sigmoid, and SoftMax functions are applied on the feature maps to make the system non-linear. With the help of these operations feature vectors are obtained which can then be utilized for tasks like classification, segmentation, and quantification of the disease as shown in fig 4. Many research works are aimed at employing the pre-trained models directly for COVID-19 diagnosis like AlexNet, Visual Geometry Group (VGG-16/19), U-Net, ResNet, Xception, and GoogLeNet. Another approach taken involves the development of new and improved DL-based architectures that are trained on the CT or Xray datasets available. For the diagnosis of COVID-19, many DL-based models have been utilized employing publically available datasets. The DL models are trained using three approaches namely supervised, unsupervised, and reinforcement learning. 

For evaluation of the Deep Learning models for the classification task, various parameters can be utilized. These evaluation parameters include confusion matrix, accuracy, specificity, recall, precision, etc [63] . The confusion matrix represents a matric of correct as well as false classifications as shown in Fig 5. The four values represented in the confusion matrix include True Positive (TP), True Negative (TN), False Positive (FP), and False Negative (FN). With the help of the values obtained from the confusion matrix, other evaluation parameters like the accuracy, specificity, recall, and precision can be further calculated [64] . The TP value specifies the number of cases that have been correctly identified as having the disease which in our case is the COVID-19 pneumonia. In other words, they have been correctly or truly classified as COVID-19 positive. The TN category includes cases that don't have the disease and have been correctly identified as normal or COVID-19 negative. The other two categories of FN and FP indicates the number of cases for which the model has misdiagnosed. FN indicates the number of cases that are falsely diagnosed as normal or COVID-19 negative but have the disease [65] . This can be very dangerous as the person will not know that he has the disease and will not get timely treatment or support and can also spread the virus leading to more cases. The other category of False Positive is falsely diagnosed to be COVID-19 positive or in general, the system predicts that they have the disease when they are normal. This again can lead to a lot of mental and physical disturbance to the person involved. So, any DL classification algorithm aims to increase the number of TP and TNs and reduce to zero the FP and FN cases.

True Positive (TP) False Negative (FN)

False Positive (FP) True Negative (TN) 

The accuracy gives a percentage of how many cases have been correctly identified i.e. TP and TN from the total number of cases. This is shown in equation 1.

As we know that TP gives the number of cases that are correctly identified as having the disease and FP gives a false classification of COVID-19 positive who don't have the disease. TP+FP will provide the number of cases that the model has identified as having the disease out of which only TP values represent correct classification. The precision value gives the percentage of TP out of the total positive cases diagnosed by the system as shown in equation 2.

It is also called as sensitivity and is represented as a ratio between the TP values and the summation of TP and FN as shown in equation 3. The FN identifies the falsely diagnosed cases as COVID-19 negative that in reality have the disease. The ratio will provide an insight into the number of cases that have been correctly diagnosed as COVID-19 positive from the total number of COVID-19 positive cases whether diagnosed correctly or incorrectly.

The F1-Score also provides a measure of the classification models' accuracy in terms of Precision and Recall as shown in equation 4.

Similarly, specificity provides a ratio between the correctly or truly diagnosed negative cases i.e. normal cases, and the summation of TN and FP cases that indicates the total number of normal cases seen by the model.

The DL models can be used for COVID-19 diagnosis, effective screening, case classification, segmentation, and quantification of the infection.

Shuai Wang et al. [66] propose the first method to effectively screen and diagnose COVID-19 patients from other viral pneumonia cases. Their method consists of three main steps: 1) Pre-processing; 2) Feature extraction of region of Interest (ROI) patches and training; Xiaowei Xu et al. [67] present a screening model to distinguish between COVID-19 pneumonia, Influenza-A viral pneumonia, and Normal cases. In their multi-center study, CT images were acquired from three COVID-19 designated hospitals in China. Out of the 618 CT images collected, 219 from 110 patients had COVID-19, 224 had Influenza-A viral pneumonia, H1N1, H3N2, H5N1, H7N9, etc, and 175 CT scans were obtained from healthy individuals. (85.4%) were used for training and validation sets and the remaining 14.6% were used as the test set. A 3D CNN model is used to segment out the pre-processed CT scan into multiple image patches. These patches are classified in the above mentioned 3 categories and confidence scores are generated. The Noisy-or Bayesian function is used to calculate the total confidence score and infection type for the entire CT scan with an accuracy of 86.7 %. Lin Li et al. [68] have proposed a method to automatically detect the effect of COVID-19 using the Chest CT scans and simultaneously distinguish it from Community-acquired pneumonia (CAP). The dataset was collected from 6 hospitals with 4356 chest CT exams from 3,322 patients. They have developed a multi-center 3D deep learning model called COVNet with ResNet50 as the backbone. Initially, a 3D CT scan is pre-processed and the Region of Interest (ROI) is extracted using the U-net model and subsequently, the COVNet is utilized to detect COVID-19. The per exam sensitivity for the test set is 114 of 127 and specificity for detecting COVID-19 is 294 of 307 with an Area Under the Curve (AUC) of 0.96.

Shuo Wang et al. [69] present a fully automatic diagnostic DL network for COVID-19 detection and survival prediction or prognosis evaluation using CT scans. They collected 5372 CT scans from 7 hospitals from 2 main datasets namely the COVID-19 dataset with 1266 scans and the CT-epidermal growth factor receptor (CT-EGFR) dataset with 4106 scans. A total of 471 cases with COVID-19 are utilized for prognostic analysis and regular follow-up for at least 5 days was undertaken. For the training set, the diagnostic performance in terms of AUC, sensitivity, and specificity is equal to 0.90, 78.93%, and 89.93% respectively. Two other external validation sets are formulated and the performance of the model is evaluated with AUC = 0.87 and 0.88, sensitivity = 80.39% and 79.35%, specificity = 76.61% and 81.16%. Xuehai He et al. [70] developed a Self-Trans approach which integrates self-supervised learning with transfer learning to achieve improved diagnostic accuracy for small datasets. They further develop a dataset containing hundreds of publically-available positive COVID-19 CT scans. Their model approaches an AUC of 0.94, an F1 score of 0.85 for detecting COVID-19 when trained with a few hundred CT scans. Shaoping Hu et al. [71] have proposed a weakly supervised deep learning method for diagnosis and classification of COVID-19. They perform binary classification of COVID-19 from non-COVID-19 cases and can detect infectious areas in CT images with high accuracy. They perform three-way classification and binary classification with two learning configurations called single-scale learning and multiscale learning. The different configurations are compared in terms of accuracy, precision, sensitivity, specificity, and AUC.

Hengyuan Kang et al. [72] develop a system to extract features from CT images and automatically diagnose COVID-19 pneumonia and distinguish it from CAP. A unified latent representation is learned to fully encode different aspects of the features. They utilized a total of 2522 CT images among which 1495 cases are from COVID-19 patients and 1027 cases are from CAP patients. With the proposed method diagnosis performance is improved with evaluation parameters like accuracy, sensitivity, and specificity of 95.5%, 96.6%, and 93. Tables 2 and 3 .

Chuansheng Zheng et al. [77] report a weakly supervised DL model called DeCoVNet for the detection of COVID-19 with weak labels using the 3D CT scans. The U-Net model is used to first segment the lung region that is given into a 3D Deep Neural Network to calculate the probability of occurrence of COVID-19. A total of 499 CT volumes was used for training the model and 131 CT volumes for testing purposes. Fei Shan et al. [79] presented a model to automatically segment and quantify the lung infection volume in CT images using Deep Learning. In total 300 COVID-19 CT images were collected from Shanghai Public Health Clinical Center for validation and 249 COVID-19 CT images were obtained from other centers that were outside Shanghai. The multi-center system employed "VB-Net" neural network to segment COVID-19 infection regions. The training was implemented using the Human-in-the-Loop (HITL) strategy with 249 COVID-19 cases, and validated using 300 new COVID-19 cases. The system is evaluated using various parameters like the Dice Coefficient (DC) equal to 91.6% ± 10.0%, Percentage of Infection (POI), and difference of volume is calculated between the manual delineation and results obtained by the model. Longxi Zhou et al. [80] have developed a multi-center model to segment and quantify the infection regions from CT scans. They collected 201 CT scans of 140 COVID-19 patients from 4 different hospitals in Heilongjiang Province, China, and 20 COVID-19 CT scans from King Faisal Specialist Hospital and Research Center (KFSHRC) in Riyadh, Saudi Arabia. They have presented a fully automatic and machine agnostic method including the development of a CT simulator to incorporate all the changes in patient's data collected at different times. Furthermore, they present a novel deep learning algorithm that decomposes a 3D segmentation problem into 3 2D problems thus reducing complexity. The segmentation performance on the Harbin dataset is evaluated using Dice Coefficient (DC), recall, and worst-case with values equal to 0.783±0.080, 0.776±0.072, and (0.577, 0) respectively. They can reduce the problem of data scarcity and further can identify small objects in a bigger scenario. Hai-tao Zhang et al. [81] present a model to diagnose, localize, and quantify the COVID-19 infection. A total of 2460 CT scans were obtained from Huoshenshan Hospital in Wuhan, China, and a uAI Intelligent Assistant Analysis System was used to assess the images. The deep learning-based software is composed of a modified 3D CNN and a V-Net that is used to analyze the infectious regions and calculate the distribution and volume of infection in the lungs. Out of the 2460 CT images, 2215 (90%) had clinical manifestations of COVID-19 pneumonia in both lungs. Table  4 presents a comparison of the techniques discussed above in terms of accuracy. Table 4 : Comparison of schemes based on the highest accuracy achieved for detection using CT images.

Highest Achieved Accuracy (%) in Classification and Detection Shuai Wang et. al. [66] 79.3%, Xiaowei Xu et al. [67] 86.7 % Hengyuan Kang et al [72] 95.5%, Vruddhi Shah et al. [74] (VGG19) 94.52%. Ali Abbasian Ardakani et al. [75] (ResNet101) 99.51%.

Zhongyi Han et al. [76] 97.9%

Many efforts have been made to detect COVID-19 with DL models most of which are built on pre-trained models. The pre-trained models have been already trained on enormous databases of images like the ImageNet dataset and can be repurposed for COVID-19 detection [82] . This is known as Transfer learning and can be extremely useful for application wherein data samples are insufficient. Several research works including pre-trained models used with Transfer Learning and Custom models are discussed below. Tulin Ozturk et al. [83] present a new DL model for detection, binary, and multi-class classification using X-ray images. They selected the Darknet-19 classifier model that forms the underlying framework for the real-time object detection model called You-Only-Look-Once (YOLO). Inspired by the Darknet architecture, they have designed the DarkCovidNet architecture which can detect as well as classify between COVID-19 vs normal vs pneumonia categories. The model can achieve an accuracy of 98.08% for binary classification and 87.02% for multi-class classification.

Nour Eldeen M. Khalifa et al. [84] propose a DL model based on Generative Adversarial Networks (GANs) and transfer learning for detection of COVID-19 using a small Chest X-ray dataset. The GAN is used to augment the dataset and prevents the model from overfitting. Although the dataset had 5863 images but only 10% of those were used for training and the rest were developed using the GAN and had only two categories: Normal and Pneumonia. They selected the pre-trained models like GoogLeNet, AlexNet, Squeeznet, and Resnet18 and further utilized transfer learning to perform the binary classification. They evaluated that Resnet18 has the most superior performance with a testing accuracy of 99.00%. [85] propose a DL model to detect and distinguish COVID-19 cases from viral pneumonia and normal cases. They fine-tuned models already trained on ImageNet and applied the transfer learning approach twice to train the model on the NIH ChestX-Ray14 dataset. Furthermore, they adopted strategies like output neuron keeping and used Layer-wise Relevance Propagation (LRP) to produce heatmaps. The model can achieve an accuracy of 99.4% and an F1 score of 0.994.

Ali Narin et al. [86] present a CNN-based model for automatic detection of COVID-19 using chest X-ray radiographs. They have evaluated three different CNN-based models like the InceptionV3, ResNet50 and, Inception-ResNetV2 for their applicability in the COVID-19 detection task. The pre-trained ResNet50 model achieves the highest classification accuracy of 98%. The other two models: InceptionV3 and Inception-ResNetV2 achieve accuracies of 97% and 87% respectively. The dataset used by them includes 224×224 sized 100 images in total with 50 having COVID-19 and 50 belonging to the normal category.

Ezz El-Din Hemdan et al. [87] have developed COVIDX-Net DL-classifier for COVID-19 detection using X-ray images. Their proposed model using pre-processing steps on the input data like scaling and one-hot encoding. Subsequently, seven different architectures of CNNs like the VGG19 and Xception, DenseNet201, and MobileNetV2 among others are utilized for accurate classification. The dataset used in this model contains 50 X-ray images that are split into training, validation, and testing set. They conclude that the VGG19 and DenseNet201 models perform the best for classification with accuracy up to 90% and InceptionV3 is identified to have the worst performance with an accuracy of 50%.

Ioannis D. Apostolopoulos et al. [88] present a transfer learning-based approach for the automatic detection of COVID-19 cases. They perform multi-class classification with 3 categories: 1) Common pneumonia, 2) COVID-19, and 3) Normal Category. The number of images in the three categories include 224 for COVID-19, 700 for Common pneumonia, and 504 for the normal category with a total number equal to 1427. They conclude that the VGG19 and MobileNet can achieve the best accuracy among the various models that also include Inception, Xception, and Inception ResNetV2. The overall accuracy of the method reaches up to 97.8% as shown in Table 5 . Harsh Panwar et al. [98] present a new DL-based model called nCOVnet which can rapidly detect COVID-19 using X-Ray images. They apply pre-processing on the original dataset like rotation, horizontal and vertical flip, and rescaling of image size to 224 × 224 × 3. The nCOVnet model has 24 layers including input and output with the remaining being a combination of Convolution, ReLU, and Max Pooling layers. They have taken into consideration the problem of data leakage and produce unbiased results. The model uses VGG-16 pre-trained on the ImageNet dataset. The overall accuracy of the model is 88.10% with sensitivity equal to 97.62% and specificity equal to 78.57%. [99] propose a method to investigate four different CNN architectures for the diagnosis of COVID-19 using Chest X-Ray images. The models include Inception V3, ResNet50, MobileNet, and Xception all of which are pre-trained on the ImageNet dataset. They found the accuracies of the Inception V3, ResNet50, MobileNet, and Xception to be 0.981, 0.825, 0.986, and 0.974. Accordingly, they conclude that the MobileNet has achieved the best performance out of the four evaluated models. [100] report a method to detect COVID-19 using the X-Ray images with the help of Grid Search (GS) and pre-trained CNN. They evaluate three different models: 1) GoogleNet; 2) ResNet18; and 3) ResNet50 for classification between 4 categories like Bacterial pneumonia, Viral pneumonia, Normal cases, and COVID-19. The combination of GS with ResNet50 yields the best performance equal to 97.69%.

Kabid Hassan Shibly et al. [101] present a Faster R-CNN framework to detect COVID-19 in patients using the X-Ray images. Their proposed method can achieve a binary classification accuracy of 97.36%, sensitivity and, precision of 97.65%, and 99.28% respectively. The custom dataset utilized by them consists of 5450 X-Ray images from the COVID Chest X-Ray dataset developed by Cohen et al. and the RSNA pneumonia dataset. They also used the COVIDx dataset with a total of 13800 X-Ray images. 

COVID-19 pandemic is an overwhelming challenge for the medical community around the globe. To tackle the effects of the outbreak AI and DL-based methods can be employed in the areas like screening, treatment, vaccine development, contact tracing and, survival prediction. In this study, we have presented an overview of the different application areas of DL and further explained in detail its use in the sub-task of detection and diagnosis of COVID-19 with the help of imaging modalities like CT-scan and X-Ray images. Several methods have been proposed that employ CNN-based models either custom-built or developed over a pre-trained model like VGG-19, ResNet50 among others. These models are not only able to detect and diagnose COVID-19 but can also classify COVID-19 pneumonia from CAP, other viral pneumonia, and the normal cases with a high level of accuracy. Furthermore, the tasks of segmentation and infection quantification can also be effectively performed with the help of the DL models. It can be concluded that the DL-based systems can assist the healthcare professional in developed as well as rural areas to mitigate the challenge of COVID-19.

Clinical features of patients infected with 2019 novel coronavirus in Wuhan, China

Correlation of Chest CT and RT-PCR Testing in Coronavirus Disease 2019 (COVID-19) in China: A Report of 1014 Cases

First cases of coronavirus disease 2019 (COVID-19) in France: surveillance, investigations and control measures

A novel coronavirus outbreak of global health concern

Early transmission dynamics in Wuhan, China, of novel coronavirusinfected pneumonia

Systematic Review and Meta-Analysis of the Efficacy of Nanoscale Materials Against Coronaviruses-Possible Potential Antiviral Agents for SARS-CoV-2

Design of RT-PCR Primers Based on Informational Method

Comparison of detection efficiency of real-time polymerase chain reaction chip for fluorescence detection by different types of coating materials

Can AI Help in Screening Viral and COVID-19 Pneumonia?

Detecting Regions at Risk for Spreading COVID-19 Using Existing Cellular Wireless Network Functionalities

Applications of machine learning and artificial intelligence for Covid-19 (SARS-CoV-2) pandemic: A review

A Review of Coronavirus Disease-2019 (COVID-19)

A Comprehensive Review of the COVID-19 Pandemic and the Role of IoT, Drones, AI, Blockchain, and 5G in Managing its Impact

A model for the effective COVID-19 identification in uncertainty environment using primary symptoms and CT scans

COVID-19 pandemic and challenges for socio-economic issues, healthcare and National Health Programs in India

COVID-19 pandemic and challenges for socio-economic issues, healthcare and National Health Programs in India

An automatic method for lung segmentation and reconstruction in chest X-ray using deep neural networks

New machine learning method for image-based diagnosis of COVID-19

Identifying Medical Diagnoses and Treatable Diseases by Image-Based Deep Learning

Computer Vision and Radiology for COVID-19 Detection

An interpretable mortality prediction model for COVID-19 patients

Artificial Intelligence and COVID-19: Deep Learning Approaches for Diagnosis and Treatment

ChestNet: A Deep Neural Network for Classification of Thoracic Diseases on Chest Radiography

CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning

COVID-19 Detection Through Transfer Learning Using Multimodal Imaging Data

A Lightweight Deep Learning Model for COVID-19 Detection

Deep Learning COVID-19 Features on CXR Using Limited Training Data Sets

Deep Learning for The Detection of COVID-19 Using Transfer Learning and Model Integration

Iteratively Pruned Deep Learning Ensembles for COVID-19 Detection in Chest X-Rays

A combined deep CNN-LSTM network for the detection of novel coronavirus (COVID-19) using X-ray images

Diagnosis and detection of infected tissue of COVID-19 patients based on lung x-ray image using convolutional neural network approaches

WHO: World Health Organization

Contact tracing in the context of COVID-19

Wikipedia: Covid-19 apps

Tracing Tracker -a flood of coronavirus apps is tracking us. Now it's time to keep track of them

Contact tracing for the control of infectious disease epidemics: chronic wasting disease in deer farms

COVID-19 Future Forecasting Using Supervised Machine Learning Models

Prediction of COVID-19 Outbreak in India adopting Bhilwara Model of Containment

Time series forecasting of COVID-19 transmission in Canada using LSTM Networks

Application of the ARIMA model on the COVID-2019 epidemic dataset

Prediction of criticality in patients with severe Covid-19 infection using three clinical features: a machine learning-based prognostic model with clinical data in Wuhan

Real-time forecasts and risk assessment of novel coronavirus (COVID-19) cases: a data-driven analysis

Short-term forecasting COVID-19 cumulative confirmed cases: perspectives for Brazil

Antibody Test/Vaccination Certification: There's an App for That

Artificial intelligence approach fighting COVID-19 with repurposing drugs

Déjà vu: stimulating open drug discovery for SARS-CoV-2

Open drug discovery for the Zika virus

Predicting commercially available antiviral drugs that may act on the novel coronavirus (SARS-CoV-2) through a drug-target interaction deep learning model

AI Techniques for COVID-19

Monitoring COVID-19 social distancing with person detection and tracking via fine-tuned YOLO v3 and deep sort techniques

Deep Learning for Classification and Localization of COVID-19 Markers in Pointof-Care Lung Ultrasound

Object Detection with Deep Learning: A Review

Comparative analysis of classification algorithms

Convolutional neural networks for image classification

Advancements in Image Classification using Convolutional Neural Network

Understanding of a convolutional neural network

COVID-19 Data Collection

A Comprehensive Survey of COVID-19 Detection using Medical Images

Performance evaluation of various parameters of Network-on-Chip (NoC) for different topologies

An Investigation on Facial Emotional Expression Recognition Based on Linear-Decision-Boundaries Classifiers Using Convolutional Neural Network for Feature Extraction

Automated Image Quality Evaluation of Structural Brain Magnetic Resonance Images using Deep Convolutional Neural Networks

CNN-based Deep Learning Model for Chest X-ray Health Classification Using TensorFlow

A deep learning algorithm using CT images to screen for Corona Virus Disease (COVID-19)

A Deep Learning System to Screen Novel Coronavirus Disease 2019 Pneumonia

Using Artificial Intelligence to Detect COVID-19 and Community-acquired Pneumonia Based on Pulmonary CT: Evaluation of the Diagnostic Accuracy

A fully automatic deep learning system for COVID-19 diagnostic and prognostic analysis

Weakly Supervised Deep Learning for COVID-19 Infection Detection and Classification from CT Images

Diagnosis of Coronavirus Disease 2019 (COVID-19) With Structured Latent Multi-View Representation Learning

Inf-Net: Automatic COVID-19 Lung Infection Segmentation from CT Scans

Diagnosis of COVID-19 using CT scan images and deep learning techniques

Application of deep learning technique to manage COVID-19 in routine clinical practice using CT images: Results of 10 convolutional neural networks

Accurate Screening of COVID-19 using Attention Based Deep 3D Multiple Instance Learning

Deep Learning-based Detection for COVID-19 from Chest CT using Weak Label

Learning spatiotemporal features with 3D convolutional networks

Lung Infection Quantification of COVID-19 in CT Images with Deep Learning

A Rapid, Accurate and Machine-Agnostic Segmentation and Quantification Method for CT-Based COVID-19 Diagnosis

Automated detection and quantification of COVID-19 pneumonia: CT imaging analysis by a deep learning-based software

Detection of Pediatric Pneumonia from Chest X-Ray Images using CNN and Transfer Learning

Automated detection of COVID-19 cases using deep neural networks with X-ray images

Detection of coronavirus (COVID-19) associated pneumonia based on generative adversarial networks and a fine-tuned deep transfer learning model using chest x-ray dataset

A deep convolutional neural network for COVID-19 detection using chest x-rays

Automatic detection of coronavirus disease (COVID-19) using x-ray images and deep convolutional neural networks

Covidx-net: A framework of deep learning classifiers to diagnose COVID-19 in x-ray images

Covid-19: automatic detection from x-ray images utilizing transfer learning with convolutional neural networks

Very Deep Convolutional Networks for Large-Scale Image Recognition

MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

Inception-v4, inception-resnet and the impact of residual connections on learning

Xception: Deep Learning with Depthwise Separable Convolutions

CoroNet: A deep neural network for detection and diagnosis of COVID-19 from chest x-ray images

Classi_cation of COVID-19 from chest x-ray images using deep convolutional neural networks," medRxiv

Detection of Coronavirus Disease (COVID-19) Based on Deep Features and Support Vector Machine

COVID-Net: A Tailored Deep Convolutional Neural Network Design for Detection of COVID-19 Cases from Chest X-Ray Images

Deep residual learning for image recognition

Application of deep learning for fast detection of COVID-19 in X-Rays using nCOVnet

Deep Learning-based Diagnosis Recommendation for COVID-19 using Chest X-Rays Images

A Deep Learning Framework for Coronavirus Disease (COVID-19) Detection in X-Ray Images

COVID faster R-CNN: A novel framework to Diagnose Novel Coronavirus Disease (COVID-19) in X-Ray images

[16] Lin L., Lixin Q., Zeguo X., Youbing Y., Xin W., Bin K., Junjie B., Yi L., Zhenghan F., Qi S., Kunlin C., Daliang L., Guisheng W., Qizhong X., Xisheng F., Shiqin Z., Juan X., and Jun X., "Using Artificial Intelligence to Detect COVID-19 and Community-acquired Pneumonia Based on Pulmonary CT:" Radiology 2020 296:2, E65-E71.