Applying Different Machine Learning Techniques for Prediction of COVID-19 Severity

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• The global increase in SARS-CoV-2 deaths necessitates technological solutions for healthcare.
• Accurate prediction of patient severity and mortality risk is crucial for timely medical intervention and resource allocation.
• Chest X-rays are utilized for early detection of COVID-19, presenting an opportunity for AI-driven analysis.

Purpose of the Study:

• To develop and evaluate a machine learning model for predicting COVID-19 patient severity using X-ray images.
• To compare the effectiveness of different feature extraction and selection methods in conjunction with various machine learning classifiers.
• To identify the optimal model for early risk stratification of COVID-19 patients.

Main Methods:

• Utilized CheXNet deep pre-trained model and hybrid handcrafted techniques for feature extraction.
• Applied Principal Component Analysis (PCA) and Recursive Feature Elimination (RFE) for feature selection.
• Trained and evaluated six machine learning classifiers, including XGBoost and Support Vector Machine (SVM).

Main Results:

• The combination of PCA and RFE for handcrafted features yielded the best results across classifiers.
• XGBoost achieved 97% accuracy, 98% precision, 95% recall, 96% F1-score, and 100% ROC-AUC with merged features.
• SVM demonstrated strong performance, achieving 97% accuracy with merged features.
• Using CheXNet features, Extra Tree and SVM classifiers with RFE attained 99.6% in all performance metrics.

Conclusions:

• Machine learning models, particularly those combining PCA+RFE feature selection with XGBoost or SVM, show high efficacy in predicting COVID-19 severity from X-rays.
• AI-powered analysis of chest X-rays offers a promising approach for early risk assessment, potentially improving patient outcomes and optimizing hospital resource management.
• The study highlights the potential of leveraging deep learning and traditional machine learning techniques for robust COVID-19 patient stratification.