Fusion-Based Deep Learning Approach for Accurate Cardiovascular Disease Diagnosis using ECG and Structured Data

Cardiovascular disease (CVD) continues to be a major global health issue with a significant burden in mortality and morbidity. Ideally, earlier and more accurate diagnosis will improve patient outcomes and reduce health system burdens. Common diagnosis methods of CVD often use specific clinical data or clinical ECG interpretations which frequently lack interpretability and reliance on expert opinion. This paper presents an advanced deep learning based system that fuses structured clinical data with ECG images to improve the accuracy of predicting heart disease. The MLP used patient history while DenseNet was applied to extract features from ECG images demonstrating a multimodal system. By using both modalities together, clinicians can be more accurate in diagnosing heart disease by moving beyond binary classification towards multi-disease classification, which eliminates many of the issues with binary classifications. A more accurate system using multimodal data will reduce false positives, and provide greater generalisability across the patient population. This research introduced a scalable and automatic system which can be adapted into clinical situations in a clinical to support earlier intervention and better data driven decision making in heart health. This framework is a major leap forward in intelligent cardiovascular diagnosis, providing a whole-of- systems solution that considers multimodal data.