Detection of Lung Affected by the COVID-19 Using New CNN Architecture

Abstract

Accurate and early detection of COVID-19-related lung infection is crucial for effective clinical decision-making. This paper presents a novel four-stage convolutional neural network (4CNN) architecture for classifying chest X-ray images into three clinically consistent categories: COVID-19 infected, non-COVID abnormal, and normal lungs. The proposed model addresses dataset imbalance using class-weighted learning and evaluates performance using comprehensive medical metrics. Experimental results demonstrate strong classification performance with an overall accuracy of 92% and improved sensitivity for minority classes. 4CNN architecture is employed to provide visual explainability.