A Deep Learning based Solution (Covi-DeteCT) Amidst COVID-19

The whole world has been severely affected due to the COVID-19 pandemic. The rapid and large-scale spread has caused immense pressure on the medical sector hence increasing the chances of false detection due to human errors and mishandling of reports. At the time of outbreaks of COVID-19, there is a crucial shortage of test kits as well. Quick diagnostic testing has become one of the main challenges. For the detection of COVID-19, many Artificial Intelligence based methodologies have been proposed, a few had suggested integration of the model on a public usable platform, but none had executed this on a working application as per our knowledge.

Keeping the above comprehension in mind, the objective is to provide an easy-to-use platform for COVID-19 identification. This work would be a contribution to the digitization of health facilities. This work is a fusion of deep learning classifiers and medical images to provide a speedy and accurate identification of the COVID-19 virus by analyzing the user's CT scan images of the lungs. It will assist healthcare workers in reducing their workload and decreasing the possibility of false detection.

In this work, various models like Resnet50V2 and Resnet101V2, an adjusted rendition of ResNet101V2 with Feature Pyramid Network, have been applied for classifying the CT scan images into the categories: normal or COVID-19 positive.

A detailed analysis of all three models' performances have been done on the SARS-CoV-2 dataset with various metrics like precision, recall, F1-score, ROC curve, etc. It was found that Resnet50V2 achieves an accuracy of 96.79%, whereas Resnet101V2 achieves an accuracy of 97.79%. An accuracy of 98.19% has been obtained by ResNet101V2 with Feature Pyramid Network. As Res- Net101V2 with Feature Pyramid Network is showing better results, thus, it is further incorporated into a working application that takes CT images as input from the user and feeds into the trained model and detects the presence of COVID-19 infection.

A mobile application integrated with the deeper variant of ResNet, i.e., ResNet101V2 with FPN checks the presence of COVID-19 in a faster and accurate manner. People can use this application on their smart mobile devices. This automated system would assist healthcare workers as well, which ultimately reduces their workload and decreases the possibility of false detection.