Res-UNet Supported Segmentation and Evaluation of COVID19 Lesion in Lung CT

Kesavan, S. M.; Al Naimi, I.; Al Attar, F.; Rajinikanth, V.; Kadry, S.

COVID19 is one of the hash lung infections;which causes severe pneumonia in humans and untreated infection will lead to death. The goal of this study is to employ an automated Infection-Segmentation-Scheme (ISS) to extract and evaluate the COVID19 lesion on CT scans of the Lungs. This work implemented a Convolution-Neural-Network (CNN) scheme called Res-UNet to study the CT slices of the lungs. The various phases of this research involve in;(i) 3D to 2D conversion and resizing, (ii) Implementation of CNN segmentation scheme, (iii) Comparison of mined COVID19 lesion with Ground-Truth (GT) and (iv) Validation. In this study, 200 CT images (10 patients x 20 slices/patient) of dimension 224× 224× 3 pixels are considered for the assessment and the Image-Quality-Measures (IQM), like Jaccard, Dice ad Accuracy are computed between extracted lesion and the GT. The experimental outcome confirms that the result of Res-UNet is better on sagittal-view of CT compared to axial and coronal. © 2021 IEEE.