Diagnosis of COVID-19 Using Deep Learning Augmented with Contour Detection on X-rays

Abstract

The WHO has declared the infectious respiratory disease COVID-19 caused due to novel coronavirus an international pandemic on March 11, 2020. The raging pandemic created a colossal loss of human life and created economic and social disruptions for millions worldwide. As the illness is new, the medical system and infrastructure are presently inadequate to counter the condition. The situation demands innovating creatively and instituting countervailing measures to circumvent the crisis. Artificial intelligence and machine learning need to be the engine for leading the technological transformation across the healthcare industry amidst the pandemic. As human resources have stretched, automation is the key to tiding over the situation. Early computer-aided automated detection of the disease can provide the necessary edge in combating this deadly virus. Though the availability of a dataset with sufficient ground truth remains a challenge, a convolution neural network (CNN)-based approach can play a dominant role as a classifier solution for the chest X-rays. This paper explores the possibility of using the convolutional neural networks to classify the chest X-rays on the preprocessed images using thresholding followed by morphological processing for edge detection and contour detection. The accuracy of a stand-alone CNN network increases remarkably when preprocessed images are used as input.