Integrating Multi-Dimensional Features for Accurate Cardiovascular Disease Diagnosis Using Advanced Fusion Techniques

Abstract

Cardiovascular diseases (CVDs) are the primary cause of death worldwide, responsible for approximately one-third of total deaths. The intricacy of CVD diagnostics necessitates the use of various biomedical signals and clinical information to attain in-depth and accurate diagnoses. Most conventional unimodal diagnostic systems fail in addressing the multi-aspect characteristics of cardiovascular conditions. In this research, a multi-dimensional feature fusion model is proposed and tested to enhance the accuracy and interpretability of CVD diagnosis. The foremost objective is to overcome issues related to data heterogeneity, computational complexity, and real-time deployment in clinical applications. A systematic review of new fusion methods was performed, and a new model combining five modalities: ECG, PPG, echocardiogram video, heart sounds, and clinical text, was built. Modality-specific neural networks were used to extract features, and feature-level fusion was carried out with concatenation and an MLP classifier. Performance was empirically verified on several public databases using accuracy, precision, recall, and F1-score performance metrics. SHAP analysis was also applied for increased interpretability. The model with the proposed architecture showed dramatic improvements in performance for all metrics of evaluation. With all modalities combined, the model reached a diagnostic accuracy of 96.8%, surpassing any partial or unimodal combination. SHAP analysis identified the relative contributions of each modality, with echocardiogram and ECG features having the most predictive power. Multi-dimensional fusion of features has a revolutionary approach to the diagnostics of CVD by successfully consolidating disparate biomedical data. Explanation and privacy-friendly models, in addition to real-time integration in wearable and remote monitoring applications, are what need to be ventured in future endeavors.