Genetically Optimized Computer-Aided Diagnosis for Detection and Classification of COVID-19

Abstract

A coronavirus called COVID-19 from China affected many people around the globe. Early diagnosis plays a major role in decreasing the death rate of COVID-19 patients. A computer-aided diagnosis (CAD) system built using Artificial Neural Network (ANN) and a fuzzy classifier will be helpful in proper and accurate diagnosis, eliminating the demerits of manual methods such as human error and inaccuracy, and reducing the time for diagnosis while cases are increasing daily. This work proposes an optimized automatic CAD system for segmentation and classification of abnormalities in CT images using genetic computations. The proposed approach extracts the statistical and wavelet features to classify the CT images as normal and abnormal using an optimized fuzzy classifier, whose fuzzy rules are optimized using a Genetic Algorithm (GA). The abnormal images are further processed for suspicious regions detection using the GA technique. Texture and intensity features are extracted from the suspicious regions and classified as pneumonia, COVID-19 pneumonia, and other infections using an ANN model, whose input features, initial weights, and hidden nodes are optimized using GA. The proposed ABCNN approach is evaluated using 470 CT lung images collected from public datasets and was shown to achieve an accuracy of 89.83%, sensitivity of 93.03%, and specificity of 83.33%.