Explainable artificial intelligence approaches for COVID-19 prognosis prediction using clinical markers

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Infectious Disease Modeling

Background:

• The COVID-19 pandemic caused significant mortality, with certain populations remaining vulnerable despite vaccination.
• Early identification of severe COVID-19 cases is crucial for timely intervention and improved patient outcomes.
• Clinical and laboratory markers offer potential for predicting disease severity.

Purpose of the Study:

• To develop and evaluate machine learning and deep learning models for early prediction of COVID-19 severity.
• To identify the most significant clinical and laboratory markers associated with severe COVID-19 outcomes.
• To enhance model interpretability using explainable artificial intelligence (XAI) techniques.

Main Methods:

• Collected COVID-19 patient data from two hospitals.
• Employed nature-inspired algorithms for feature selection to identify crucial predictive markers.
• Trained and tested multiple machine learning classifiers for severity prediction.
• Utilized five explainable artificial intelligence (XAI) techniques to interpret model predictions.

Main Results:

• Achieved a maximum testing accuracy of 95% in predicting COVID-19 severity.
• XAI analysis highlighted c-reactive protein, basophils, lymphocytes, albumin, D-Dimer, and neutrophils as key predictive markers.
• The developed models demonstrate high efficacy in identifying patients at risk of severe disease.

Conclusions:

• Machine learning models can effectively predict COVID-19 severity using clinical and laboratory data.
• Explainable AI provides insights into the critical factors influencing severity predictions.
• The proposed computer-aided diagnostic method can support clinical decision-making and alleviate healthcare system burdens.