A Robust and Computationally Faster Approach to COVID-19 Diagnosis using Shallow Convolutional Neural Architecture

ABSTRACT

The ongoing COVID-19 pandemic has infected millions of people worldwide, overwhelming health infrastructures. The common symptoms are fever, cough, sore throat, muscle pain, headache, nausea, vomiting, and diarrhea, similar to the symptoms of common flu in mild and moderate cases. The distinguishing signs of common flu from that of COVID-19 are invisible until patients start feeling shortness of breath when the infection attacks the respiratory system in severe cases. At this stage, patients require immediate medical attention and hospitalization. In developing countries where health facilities are not adequate and costlier radiological tests like computed tomography (CT)-scans are scarce, diagnosis becomes a challenging task. The clarity comes only when a COVID- 19 test is conducted, which has its own time limitations depriving the patient of specialized treatment until the patient tests positive. In far to reach rural areas identification of COVID-19 induced pneumonia cases with the help of chest X-rays is more difficult with substandard medical infrastructure and a handful of expert radiologists available. This work detects COVID-19 induced Pneumonia with the help of chest X-rays. The used dataset includes COVID-19-infected patients' chest X-ray images as well as normal non-COVID chest X-Ray images. A Lightweight Stacked Convolutional Neural Network was created to extract fine details and information from images, assisting in the detection of COVID-19 caused pneumonia cases. For evaluation, we assessed the developed Neural Network model on a test validation set consisting of hundreds of chest X-Ray images. The suggested Neural Network's average test accuracy was determined to be 98.76%, with per-class accuracy of 99.40% for detecting COVID-19 cases and 98.42% for detecting normal cases.