An Explainable Decision Support Framework for Differential Diagnosis Between Mild COVID-19 and Other Similar Influenzas

Abstract

It is tough to clinically differentiate between mild COVID-19 and other similar influenzas due to their comparable transmission traits and symptoms. The Real-time reverse transcriptase-polymerase chain reaction (RT-PCR) test is utilized regularly to diagnose severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) despite being prone to false-negative results. In recent years, intelligent support systems have been developed for patient triage and disease diagnosis. Thus, this research utilizes machine learning to diagnose COVID-19 from routine biomarkers. Twelve feature selection techniques, which include nature-inspired techniques, have been compared to extract the essential features. Multiple classifiers, including stacking, voting and deep learning, are trained to predict the patient diagnosis. The maximum accuracy obtained by the classifiers was 95% in this retrospective study. The diagnostic predictions were further interpreted using five explainable artificial intelligence methods. Biomarkers such as albumin, protein, eosinophil and total white blood cells were crucial. Thus, automated diagnostic systems can be supportive in the accurate and timely detection of COVID-19 and similar influenza infections.

Keywords

• Biomarkers
• explainable artificial intelligence
• Mild COVID-19
• nature-inspired algorithms
• non-COVID-19 influenza