A Novel Deep Learning-based Approach for Covid-19 Infection Identification in Chest X-ray Image using Improved Image Segmentation Technique

Abstract

Covid-19 diagnosis systems are being improved with the emerging development of deep learning techniques. Covid-19 is widely known for the deadly effects and its high transmission rate. To overcome the challenges, different deep learning-based detection methods have been introduced through which the disease can easily be identified in patient's body. But only identification of the disease is not sufficient to assist physicians for further diagnosis. Infection identification process with severity measurement from medical image can put an advancement in current Covid-19 diagnosis systems. This work presents a novel infection detection approach based on image segmentation technique that can be used to localize the infection. The proposed system is able to predict segmented lung and mask images with visual representation so that it makes the diagnosis task easier for the physicians. ResNet-U-N et, VGG16-U-Net and a modified U-Net model have been implemented in the proposed work where the modified U-Net performed better with 0.968 IoU, 98.60% accuracy and 0.9567 of dice coefficient. An advanced module using OpenCV has been designed that can calculate the area of the predicted lung and infection mask images separately and then the infection percentage can be calculated accurately.