An Explainable Quality-Aware ECG-PCG Fusion for Cardiovascular Disease Detection Using Robust Feature Modeling

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Cardiac abnormality detection traditionally relies on expert interpretation of electrocardiogram (ECG) and phonocardiogram (PCG) signals.
• Automated methods are needed for faster and more accessible cardiac screening.

Purpose of the Study:

• To develop a rapid, interpretable, multimodal framework for automatic cardiac abnormality detection.
• To integrate ECG and PCG signals for enhanced diagnostic accuracy.

Main Methods:

• A multimodal classification scheme combining ECG and PCG recordings.
• Tailored preprocessing and feature extraction for each signal modality.
• Quality-aware fusion strategy prioritizing reliable signal segments.
• Explainability analysis to ensure focus on physiologically meaningful regions.

Main Results:

• Achieved high performance metrics: 97.2% F1 score, 95.6% accuracy, 88.6% specificity, and 97.7% sensitivity.
• Demonstrated computational efficiency, completing 10-fold cross-validation in 66 seconds.
• Explainability results confirmed focus on relevant physiological areas.

Conclusions:

• The developed ECG-PCG framework offers a fast, accurate, and interpretable solution for automated cardiac abnormality detection.
• The system is suitable for real-time cardiac screening applications.
• Multimodal signal integration enhances the reliability of automated cardiac diagnostics.