A Convolutional Neural Network Model for Screening COVID-19 Patients Based on CT Scan Images

The novel coronavirus (COVID-19) spread all over the world within a few months and turned into a pandemic. Early diagnosis is the only way to combat this pandemic by isolating the affected cases from healthy ones for refraining it from further spreading. At present, RT-PCR is extensively used for screening coronavirus cases, however, WHO stated that it suffers from low sensitivity and low specificity in the early-stage patients. Recent studies advise that the CT scan image embraces key features for detecting this disease. The application of deep learning approaches combined with CT images could be useful for the precise diagnosis of positive coronavirus patients. In this research, we have employed the Convolutional Neural Network (CNN) architecture of deep learning on publicly accessible CT images dataset to build a prediction model for classifying positive COVID-19 from other pulmonary diseases and healthy patients. Moreover, this prediction model has also been utilized for identifying COVID-19 cases from other pulmonary diseases, which is a binary classification. In ternary classification, our proposed CNN model has obtained an accuracy of 98.79%, a precision of 94.98%, a sensitivity of 98.85%. In contrast, for binary classification, it has acquired an accuracy of 98.79%, a precision of 94.98%, a sensitivity of 98.85%. Therefore, this proposed model can be a faster and alternative tool or even a supportive tool along with RT-PCR in rural areas of many countries where there is a scarcity of testing kits and expert physicians.