Deep Learning Based Classification of Covid-19 Lung Images

Abstract

COVID-19 is spreading over the entire world very faster right from the detection of the first case in china during December 2019. In order to identify this viral disease, accurate automatic quantification of the lobes in the lung by means of x-ray and CT. Automated segmentation techniques are needed to overcome the challenges like high variation in abnormal characteristics and low intensity contrast in X-ray and CT slices abnormal and normal tissues. Manual delineation is time consuming and there is a probability of having an intra-observer and inter-observer variability. Hybrid segmentation methods such as FAKM-DRLSE and K-MLRCV methods were proposed to segment the lobes in the lungs. The FAKM clustering method is used to locate the lobes which are further segmented by DRLSE method. In the second proposed method, the edge transformed image obtained from Kirsch operator is provided to the MLRCV method for segmenting the lobes. The segmentation evaluation is done with ground truth images using different evaluation metrics. The segmented images are then given to different classifiers to classify the abnormal images which are COVID infected from the normal healthy images that are non COVID affected. Dense Neural Network (DNN) gives better classification accuracy when compared to all other classifiers.