A Novel Deep Learning Approach for Classification of COVID-19 Images

Novel coronavirus 2019 (COVID-2019) initially started at Wuhan, China and spread all over the world and announced as pandemic by World Health Organization in March 2020. It makes use of all the available resources to reduce the disastrous effect of such Black Swan event. This virus causes pneumonia in human being and changes the respiratory pattern (different from common cold and flu). Compared to the reverse-transcription polymerase chain reaction (RT-PCR) chest X-ray and computed tomography imaging may be reliable and quick to diagnose the COVID-19 patients in the epidemic regions. Above mentioned imaging modalities along with the machine learning techniques can be helpful for accurate diagnosis of the disease and may be assistive in the absence of specialized physicians. Further, ML can improve throughput by accurate figuration of contagious X-ray and CT images for disease diagnosis, tracking and prognosis.In this research, potential of artificial intelligence has been investigated to develop a deep neural network model for rapid, accurate and effective COVID-19 detection from the CT and X-ray images. The proposed method provides robust deep learning technique for binary (COVID vs. NON-COVID) and multi-class (COVID vs. NON-COVID vs. Pneumonia) classification from X-ray and CT images. A 24-layer CNN network has been proposed for the classification. It attains an accuracy of 99.68% and 71.81% on X-ray and CT images, respectively. For both the datasets Sgdm optimizer has been used with a learning rate 0.001.