Deep Learning Method for Quantitative Analysis of Covid -19 using CT Chest Images

The manual diagnostic tests performed in laboratories for pandemic disease such as COVID19 is time-consuming, requires skills and expertise of the performer to yield accurate results. Moreover, it is very cost ineffective as the cost of test kits is high and also requires well-equipped labs to conduct them. Thus, other means of diagnosing the patients with presence of SARS-COV2 (the virus responsible for COVID19) must be explored. A radiography method like chest CT images is one such means that can be utilized for diagnosis of COVID19. The radio-graphical changes observed in CT images of COVID19 patient helps in developing a deep learning-based method for extraction of graphical features which are then used for automated diagnosis of the disease ahead of laboratory-based testing. The proposed work suggests an Artificial Intelligence (AI) based technique for rapid diagnosis of COVID19 from given volumetric CT images of patient’s chest by extracting its visual features and then using these features in the deep learning module. The proposed convolutional neural network is deployed for classifying the infectious and non-infectious SARS-COV2 subjects. The proposed network utilizes 746 chests scanned CT images of which 349 images belong to COVID19 positive cases while remaining 397 belong negative cases of COVID19. The extensive experiment has been completed with the accuracy of 98.4 %, sensitivity of 98.5 %, the specificity of 98.3 %, the precision of 97.1 %, F1score of 97.8 %. The obtained result shows the outstanding performance for classification of infectious and non-infectious for COVID19 cases.


INTRODUCTION
The novel coronavirus first encountered at Wuhan, China has been designated as SARS-COV2 i.e. severe acute respiratory syndrome Coronavirus 2, while the disease it causes is termed as COVID19 (Corona Virus Disease 2019) [1-2]. Coronavirus is a family of various viruses which affects humans, animals or both. In humans, various coronaviruses can cause diseases ranging from an acute illness like common cold to severe illnesses like SARS (Severe Acute Respiratory Syndrome) and MERS (Middle East Respiratory Syndrome) [3]. SARS-COV2 has "club-shaped" projection overs its spherical envelope which is also called spike glycoprotein and also possesses positive sense ssRNA (Single-stranded RNA) within its nucleocapsid. Coronaviruses are known to have 26kb -32 kb genomes which are the largest genome for RNA dominant viruses while their G+C contents range 32 to 43%. These genomes consist of a unique N-terminal fragment inside its spike and various other types of features as mentioned in [4][5]. SARS and MERS viruses namely SARS-CoV and MERS-CoV respectively are known to cause symptoms like cough, fever and breathlessness in its patients. Patients with SARS can also be effects by severe respiratory problems and are prone to a higher fatality rate as compared to other coronaviruses. Various research suggests that the origin of MERS-Cov and SARS-CoV is from bats but spread to human from different medium like SARS-CoV spread to human from civets whereas MERS-CoV spread from dromedaries. Likewise, SARS-CoV2 is also speculated to be spread from bats while the source of spread is still unclear [6].
The outbreak of uncommon and human-to-human contagious respiratory illness also call COVID19 caused by SARS-CoV2 has been termed as a pandemic by WHO (World Health Organization) as it has affected a total of 210 countries with a majority of cases from the USA, Italy, Russia, United Kingdom, China, Brazil, India etc. metric, on average, an infectious human can spread the virus to about 3 other healthy humans [3,7]. Thus, it is important to diagnose the patients suffering from COVID19 and then quarantine them for isolated monitoring and treatment. It is also observed that patients with compromised immunity due to medical history have higher fatality rate to the virus than other patients which makes early diagnosis and isolation more crucial to safeguard them especially the cancer patients who have twice more risk of getting infected than the normal person.
The patients with COVID19 can be diagnosed using various criteria i.e., symptoms, epidemiology, CT images and pathology test. The symptoms of COVID19 contains cold, cough, fever, pneumonia and other respiratory discomforts [9][10][11]. However, the above-mentioned symptoms might not be the specific COVID19 symptoms as there are many reports of asymptomatic patients tested positive through chest CT images as well as pathological test. A person under investigation (PUI) is subjected to laboratory-based pathological test by extracting a sample like bronchoalveolar lavage, sputum or tracheal aspirate from the lower respiratory system. This laboratory test is an RT-PCR(Reverse transcription-polymerase chain reaction) and nucleic acid sequencing of the virus' RNA [12] and since the outbreak is observed, the diagnosis against the COVID19 has been affected by various factors like availability, accuracy, quality, efficiency as well as the cost of these testing kits that have caused the tremendous increase in the spread of the virus as an infectious person may observe the symptoms of the virus two weeks after getting infected while he/she can still spread it to people coming in contact [13]. Moreover, the accuracy of the virus depends upon nucleic acid extraction methods, disease stages, sample collection, and on the expertise of the sample collector or human errors.
The rate of detection of nucleic acid is found to be in range to 30-50% [14] and thus repetitive testing is done to confirm the diagnosis. The radiological methods like CT scan can also be utilized for diagnosing COVID19 with more accuracy by utilizing the CT image features such as peripheral lung distribution and lung morphology, or other features like pulmonary consolidation, ground-glass opacities from late and early stage of COVID19 respectively [15].
However, there are also possibilities of these CT images being similar to patients of other infectious diseases like pneumonia [16][17][18].
Artificial Intelligence-based techniques like deep learning are developed to deal with problems like feature extraction and classification by utilizing medical images. These features may include shape-based or spatial features which are efficiently used to distinguish between the infectious and non-infectious individuals. There are various numbers of other features that can also be used for detecting pathogens by pattern analysis from input image associated with the pathogenesis [21]. CNN (Convolutional Neural network) is one such deep learning-based network which has been used to enhance the low-intensity images from given video endoscopy even with a small dataset of 55 videos [19].
CNN can also detect pneumonia using X-ray images, the nature of pulmonary swellings using CT images, or cyst from endoscopic videos [17][18][19][20]. The chest CT images of COVID19 patients in their early stages are observed to contain ground glass-opacity in one of the lungs which later progresses to both of them [22]. Similar features can also be observed in the pneumonic patients with little differences which are difficult to identify by the radiologist but can be observed by using artificial intelligence techniques like CNN.
The RT-PCR standard diagnosis was initially reported with 30-70% sensitivity while at that time the CT diagnosis was more sensitive but the recent data from US labs from the University of Washington suggests that the next generation COVID19 RT-PCR diagnosis 95 % or more sensitivity. Even after being more sensitive than RT-PCR in earlier days of the outbreak, CT images were never used for COVID19 diagnosis. Various US-based radiology society has suggested to use CT cautiously and only during the time when the diagnosis is harder to manage through pathology while there will certainly be cases where CT imaging is needed for other reasons but the presence of SARS-CoV2 is observed [1-2]. Thus, healthcare professionals need to be familiar with various imaging features of the infection.
Usually, a patient with symptoms related to COVID-19 such as shortness of breath, fever or cough is made to take chest X-RAY. The most common abnormalities of such patients involve ground glass opacities which look like frosted glass. However, it is observed that the X-rays are not much sensitive to COVID19 and can result in false-negative diagnosis whereas CT images of the chest are more detailed and distinguishable as compared to X-ray images. The common abnormalities in case of CT images are ground-glass opacities scattered throughout the lungs that are observed due to liquid filling of alveoli. In severe infections, more fluid accumulates inside the lobes of the lungs making the ground-glass appear like solid white consolidation which in certain cases is further observed as "crazy paving" pattern formed due to swelling of lung lobules present along with the interstitial space. These results in walls appear similar to irregularly shaped stones used to pave a street, hence the name "crazy paving" [21][22]. The three CT observations i.e., crazy paving patterns, consolidations and ground-glass opacities may be detected in isolation or combination with one another as shown in Fig. 1. Usually, ground glass is first to sign to appear and later followed by one or both of the other occurring in multiple lobes inside both lungs in severe patients while in mild infections the ground glass opacity can be present within one lobe. Thus, it can be said that the severity of the disease depends upon the proportion of lung findings which may or may not be observed in COVID19 patient. Moreover, the recovery of the patient can also be monitored in quantitative manner through CT imaging which is not provided by any other test. Some other findings like pleural effusions or fluid inside the lungs can also be observed in CT imaging, which is common in patients with bacterial pneumonia and congestive heart failure [1]. These observations in CT imaging namely crazy paving patterns, consolidations and ground-glass opacities can also be found in various types of viral pneumonia-like influenza and adenovirus, and other non-infectious illnesses. This implies that CT is not specifically sensitive to COVID19 and thus patients with these findings should get further clinical evaluation and laboratory test done to confirm the presence of the virus. It is recommended to utilize the CT imaging in case of unavailability of pathological testing as it can correctly identify the true negative cases but for confirming true positive cases it needs further testing. Individuals presenting to hospital with above-mentioned chest findings may need  After the study of past statistics [1-2] and Fig. 2, it is observed that in rare cases, COVID-19 can lead to severe respiratory problems, kidney failure or death. It is also observed that the findings CT images are not specific to COVID19 only and there are some cases which are suspected as COVID-19 but have a pattern on CTE that is consistent with acute lung injury, various types of viral pneumonia (influenza A, influenza B, RSV, Coronavirus, rhinovirus, parainfluenza, and adenovirus), some atypical bacteria types of pneumonia (legionella, mycoplasma), drug-induced acute lung injury (chemotherapy, immunotherapy) and cryptogenic organizing pneumonia which is an idiopathic form of primary organizing pneumonia. Table 1 explains the basic CT finding characteristics. There are some early CT findings related to viral pneumonia which appear to be dominated by bilateral peripheral ground-glass opacities. These ground-glass opacities possess round morphology which is not due to swellings or nodules found in the lungs. The consolidation can be superimposed on the ground glass opacities, but early CT findings do not contain many cases of consolidation as compared to ground-glass opacities, also in early cases discrete nodules, pleural effusion and lymphadenopathy are not reported. Among these, the pleural effusions and lymphadenopathy are not characteristic of the disease. With the progress of the disease, the ground-glass opacity becomes more confluent and often remains denser in lungs periphery but is also present throughout the lungs, in severe cases crazy paving is observed. There's also a report of linear opacities developing later in the disease process and those linear opacities being associated with architectural distortion. Also, within the areas of opacification, bronchiolar dilation is observed that may not be anywhere else in the lung [23]. The proportion of mild and asymptomatic cases versus acute and lethal cases is shown in Fig. 3. The present work aims to study the results yielded by implementing deep learning-based CNN for detecting the presence of SARS-COV2 from given CT images. The present implementation uses a total of 349 images containing both the COVID19 as well as viral pneumonia cases. The remainder sections of the paper consist of following sections: "Literature review" which describes the current literature associated with COVID19, "Proposed model" section offers a detailed structure of the proposed classification model, "Performance analysis" focuses on the analyses of results obtained in comparison to the global standards and finally "Conclusion" section contains the statements which can be inferred from the present work.

LITERATURE REVIEW
The preparedness for COVID19 in countries with exponentially increasing cases have emphasized the need for quick diagnosis, containment as well as contact tracing [24]. As the contiguousness of virus is extremely high, the nucleic The work conducted in [30] involved 1014 patients who were subjected to both types of testing i.e., RT-PCR and CT, the sensitivity related to CT imaging is found to be 97% of RT-PCR positive cases. The 81% of patients who were diagnosed negative through RT-PCR but found positive through CT were reassigned as "probable" by analyzing their clinical symptoms. The differences in bacterial or other viral pneumonia and COVID19 related pneumonia are being currently studied [31]. The crazy paving pattern is likely to be observed within 4-14 days period after the symptoms are experienced by the patients while other observations like large volume lymphadenopathy, cavitation, cysts, nodules and tree-in-bud pattern, are rarely found. CT imaging is certain to become important for diagnosing the presence of COVID19 related pneumonia in patients. Moreover, the evolution of pneumonia in patients with COVID19 along with their recovery can be easily observed through regular monitoring of CT images 2 days after the symptoms are reported. It is also reported that patients with no presence of COVID19 in CT images taken a week after experiencing symptoms are unlikely to develop pneumonia related to the virus. As the pandemic is progressing, various new data and studies are being put forward from different affected regions on a daily basis which can help in knowing other features which are specific to COVID19.
The British Society of Thoracic Imaging issued guidelines for the assessment of COVID19 suspected cases through laboratory, clinical as well as radiography and advised to reserve CT imaging for severe patients and during diagnostic uncertainty. It is also reported that in certain cases the sensitivity of CT is found to be higher than RT-PCR. RT-PCR related test requires deep throat swab or throat wash sample for detecting SARS-CoV2 and is a state-of-art testing method for the virus whereas CT imaging provides results that are suggestive for further diagnosis. The CT morphology that depends upon various features of COVID19 patients can be helpful in diagnosis especially in the case of false-negative result yielded by RT-PCR [31]. The author in [32] measured various features associated with COVID-19 from a dataset containing CT images of infectious and non-infectious patients from the place Iran and draws the hypothesis that some CT features have short-term predictive value. Pleural effusions and sub-pleural sparing are observed in one-sixth and one-fifth of COVID-19 positive patients respectively. They Studied the initial group of Iranian patients from Kashan, a hotspot of COVID-19 in Iran and also examined the features related to COVID-19 based pneumonia. The above-mentioned study [32] was conducted on low-dose CT imaging protocol designed for potential COVID-19 based pneumonia.
A study [31] indicated that peripheral distribution, GGO and vascular thickening are among most distinctive CT imaging features related to COVID-19 and necessary negative features involved in the study are hilar/mediastinal lymphadenopathy, cavitation and lack of tree-in-bud nodules. A study [32] indicates the presence of mild splenomegaly in various viral diseases which is not mentioned in earlier studies of COVID-19. There were 6 patients with normal initial CT image who were asymptomatic and came in contact with confirmed COVID-19 patients. This data emphasizes the need for using RT-PCR along with CT imaging for proper monitoring of patients' condition or recovery. The patients with severe conditions or those who lost their lives because of pneumonia related to COVID-19 are found to be considerably older and their CT findings had more crazy-paving and consolidation pattern, the involvement of peri-bronchovascular, pleural effusion and developed air bronchograms. Thus, the radiology-based feature can be utilized to describe prognostic imaging biomarkers of pneumonia related to COVID-19.
The unstable or severe patients also show involvement of peri-bronchovascular along with the subpleural distribution found in the periphery of the lungs while the stable patients are found to show perilobular opacity and RHS (reverse halo sign) [32][33][34]. Moreover, the severe, as well as the deceased patients, were found to have a higher frequency of pleural effusion, though these findings are non-COVID-19 specific and were reported by [35]. The time of taking CT scan and severity of disease are the factors affecting the findings of radiology, it is reported that in mild or acute symptomatic patients about 18% of total patients had normal CT images while only 3% of total patients had normal initial findings that later developed severe illness [36][37][38]. A study [37] described radiological findings in subclinical disease showing GGO in CT images of 15 patients taken before the symptoms are encountered. The sudden surge of COVID-19 cases in Hubei province of China on 12 February 2020 led to a shortage of testing kits and also due to time constraint, CT imaging was found to be a reliable and efficient method for diagnosis in such scenario. This helped in taking timely measures to contain the disease and treat the patients; however, the possibility of false-negative due to CT imaging taken in the early stage of the disease cannot be ignored [39][40].
A study [41] observed the rate of misdiagnosis of COVID-19 by the radiologists and found that in a sample of 51 patients, the CT features of COVID-19 were overlapping with that of infection caused by adenovirus. These overlapping involved similarities as well as dissimilarities with the CT imaging-based features of SARS (severe acute respiratory syndrome). The false-negative cases recorded through CT scan were found to be just 3.9% (2/51) and suggests CT imaging as a potential method for testing. A study [42] conducted for pediatric patients used CT as well as the laboratory-based test of 20 such patients of which 65% had contact with already positive COVID-19 family members, cough and fever were common symptoms recorded by 65% and 60% of them respectively. The features related to CT scans involved coinfection in 40%, pulmonary lesions in 30%, bilateral pulmonary lesions in 50%, 20% with no findings and 80% with procalcitonin elevation (not common among adults). A deep-learning-based technique called COVNet (COVID19 detection neural network) had been developed for extracting the features from CT images of patient's chest and then used for diagnosing the presence of COVID19 in that patient [43][44].
The CT images of non-pneumonia and CAP (Community-acquired pneumonia) are analyzed for determining the effectiveness of the model. It involves dataset taken from 6 hospitals and performance analysis of the model is done through specificity, sensitivity and AUC (Area Under curve) associated with operating characteristics of the receiver [43]. CT images that can be utilized for diagnosing the potential patients more efficiently and reliably.

MATERIAL AND METHODS
Accurate detection of COVID19 is a challenging task for all the experts of the medical fraternity. Therefore, in this work a computer assisted system has been proposed for analysis and classification of COVID19 cases using CT images and deep learning method. The description of dataset, proposed model, experiments and results are discussed here.

Database description
CT scans can prove to be vital mode of COVID19 diagnosis to provide fast, accurate and cheap and more feasible way for its screening and testing. In this paper, a dataset used for implementing the proposed technique contains 746 chest CT images taken from 143 patients with average age of 49±15 years and 2 classes namely (1) class 1: infectious i.e., CT images having positive report for COVID19 and (2) class 2: non-infectious i.e., CT images having negative report for COVID19. A total of 480 images of given dataset were used for training set, 80 images of given dataset is used as a validation set while 186 remaining images were utilized for testing set. The data is available at https://github.com/UCSD-AI4H/COVID-CT [46].

Overview of the proposed architecture
The proposed work consists of three major sections as pre-processing section, model building section and decisionmaking section. Each section consists of a set of processes and each process is very helpful in getting the desired results. The experimental workflow diagram of the proposed system is shown in Fig. 4.
Step involved in proposed work is given as: Step 1 Load input CT scan images.
Step 2 Resize all the images into 224 × 224 × 3 for similar size and equal to input layer size of proposed deep learning model.
Step 3 Remove the patient information from input CT images and reduce the noises if there is any kind of noise or artifact.
Step 4 All preprocessed images are divided into training, validation and testing set in the ratio of 65:10:25.
Step 5 Training set is applied on defined model with stochastic gradient descent kernel function and different combination of defined Identity Blocks i.e., IB-1 & IB-2.
Step 6 On the trained model, testing set is applied, and results are obtained.
Step 7 Lastly, performance evaluating parameters are computed for total testing set.

Pre-processing Section
The first section of the designed model is the preprocessing section, in which the input CT scan images are loaded.
The CT scan images of collected dataset are generated from different types of CT scanners having different features, due to this there are so many variations in image resolution. It is also known that the variations in input data size can be very big constraint for the development of any high performing system. Therefore, all input images are resized to 224 × 224 pixels. Another important part of preprocessing section is de-noising, under which each resized image is passed through the de-noising filter to remove some artifacts and noise from the input image. In every CT image, there is some important patients' information which is not useful for the desired task and using or sharing of such information is also against the ethical laws. So, this information is removed before using the images. At the end of this step, input images are in identical size, free from noise and patient's information (like his/her name, age, sex, date of scanning, time etc). So, the complete dataset is ready to use and passed to sample bifurcation.  After the training and testing set creation, normalization is performed. The major benefit of normalization is scaling and centering the data in the range of 0 and 1. It also improves the performance of the system and reduces the computation time. The normalized training and testing set is used for classification model building.

Model Building Section
The proposed model consists of two identity blocks named as IB-1 and IB-2 [47]. Each IB-X where, X= {1, 2} block consists of two paths one is called main path and other is called shortcut path. The structure of IB-1 and IB-2 are shown in Fig. 5.
The main path of IB-1 consists of three CONV_2D layers, three normalization layers and two non-linear activation functions 'ReLu'. Initially, Xin is passed through the main path and the output of main path is lastly added with original Xin using shortcut path and passed to the 'ReLu' for better performance. The IB-2 is similar to the IB-1 except the shortcut path. This shortcut path is composed of CONV_2D and normalization layers. After that Xout is obtained using addition of main path output and shortcut path output. The mathematical expression of IB-1 and IB-2 output is given in Eq.
(1) and Eq. (2). The proposed deep neural network model is comprised of stack of layers and information flow between layers for the discrimination of infectious and non-infectious cases is shown in Fig. 6.

Fig. 6. Layer Wise Data Flow in Designed Model
The detailed description and architecture of the implemented model is given is Fig. 7.

Optimized gradient descent
In past studies, it has been found that Optimized Gradient Descent or Stochastic Gradient Descent (SGD) is frequently used for DL model training and attain the promising result [48]. SGD is an optimization technique which is mathematically defined as given expression in Eq. 3 for training sample Xin with label Yin.
The algorithm used for SGD training is described in Algorithm 1.

Decision making section
In the proposed model, level 1 to level 5 are used for feature learning purpose. After feature learning, classification is performed using the average pooling layer, flatten layer, FC layer and soft-max layer. In the classification section, the instance of testing set is passed to the validated model and then obtained class label is considered as the decision of the proposed system for the particular input test sample. In the same manner, the whole test sample is passed to the validated model that assigned a label to each input sample. The performance of the model is evaluated using performance metrics, which is described in the next section.

Performance metrics
The performance of proposed work is evaluated with the help of classification accuracy, sensitivity, specificity, precision and F1score [49][50]. All parameters are obtained by the help of confusion matrix. The structure of confusion matrix (CM) is shown in Fig. 8. The mathematical expression of used parameters is given in Eq. 5, Eq. 6, Eq. 7, Eq.

EXPERIMENT AND RESULT
In this paper, extensive work has been carried for the analysis and characterization of COVID19 using CT chest images. To achieve desired outcome, deep neural network with optimized gradient descent has been used. The environmental setup for experiment execution and obtained results are discussed in next section.

Experimental Setup
The entire experimentation is performed using HP Z4 G4 workstation. This system has the below mentioned specification: Intel Xeon W-2014 CPU @ 3.2 GHz, 2TB SATA HDD and 256 GB SSD, 4GB NVIDIA Quadro P1000 and 64 GB RAM. All the images and ROIs are stored in this system and the experiments are performed using the Python environment.

Experiment and Result
In this experiment an image dataset of 746 CT scan is used.  The performance of experimentation done on the designed model is calculated in the form of classification accuracy (Class_Acc), specificity (Spe), sensitivity (Sen), precision (Prec) and F1score using previously mentioned equations.
The obtained value of each parameter is shown in Fig. 9.

Result Analysis
The outcome of the proposed model is further analyzed by two different approaches: quantitative and statistical analysis. Under quantitative analysis, accuracy, sensitivity, specificity, precision and F1score is computed, and Cohen's kappa analysis is used for statistical analysis.

Quantitative Analysis
After the regressive work, the obtained results of 186 testing samples are reported in

Statistical Analysis
The statistical analysis is performed with Cohen's kappa coefficient [51]. It shows the significance or reliability of the Where  is kappa value, P0 is the probability of observed agreement and pe is the probability of hypothetical With the help of confusion matrix given in

Comparative Analysis
The present work is directly compared with the study performed by J. Zhao et al., 2020 [46] because both are using the same set of images. The same work is also compared with the study performed by different authors [43,45,46,52]. The comparative analysis is presented in Table 4 and Fig. 13. From Table 4, it has been clearly seen that the proposed work performs better in every aspect with respect to studies [43,45,46,52]. The classification accuracy of the proposed work is superior among previously published work, which makes the system more stable. In the same manner, proposed work yields better sensitivity value, specificity value, precision and F1score. Thus, it can be concluded that the proposed work is better than the previously published work.

CONCLUSION
The majority of the deaths due to COVID19 are observed when it causes pneumonia, which itself is a huge threat for lives of general public due to its rapidly growing and fatal nature as well as its high incidence. Therefore, initial and accurate diagnosis of the ailment using every available and feasible technology is must for avoiding the grim consequences of the pandemic. The present work focused on one of such technology i.e., chest CT scans, which proved to prompt feasible and highly sensitive for such task. The proposed AI-based approach for the detection of COVID19 shows promising results with the high accuracy of 98.4 %, precision of 97.1 %, sensitivity of 98.5 %, the specificity of 98.3 % and F1 score of 97.8 %. The database used about 746 chests scanned CT images containing both infected subjects with COVID19 and normal subjects too. The results obtained by utilizing various COVID19 features of CT imaging shows that artificial intelligence-based techniques can be utilized for diagnosing the potential patients more efficiently and reliably.
The logical conclusion seems to indicate that it is hard to overcome the present COVID19 pandemic due to the highly contagious and mutating nature of its corresponding virus i.e., SARS-CoV2 but it can be prevented using the lessons learned about the preventive measures from other pandemics like MERS-CoV or SARS-CoV. This preventive measure includes using facial covers or masks, personal hygiene, maintaining social distancing and proper isolation or quarantine for potential patients of the disease. Moreover, the early diagnosis of such potential patients using whichever technique available in the area as resources are always overloaded during the pandemic. Such measures are crucial until every person in the world is completely vaccinated, as due to high mutating nature of the virus, the possibility of virus mutating to vaccine resistant variant cannot be denied. Finally, the future scope would focus on developing a robust system for AI-based diagnosis, monitoring and data collection to enable the future generation to have ample resources to deal with any pandemic.    Obtained parametric value Figure 10 Performance Curve for Training vs. Validation Loss Figure 11 Performance Curve for Training vs. Validation Accuracy Figure 12 Quantitative Analysis of Proposed Work Figure 13 Comparative Analysis between Proposed Works with State-of-the-art.