Multimodal Fusion Framework for Enhanced Heart Disease Detection Using ECG Signals and Demographic Intelligence

Abstract

Cardiovascular diseases are the greatest cause of mortality in the world, and around 17.9 million people die at the end of every year [WHO, 2023]. The present research is a Multimodal Fusion Framework (MFF-HD) that combines ECG signal processing and demographic information to detect coronary artery disease (CAD). With the help of the PTB-XL dataset (n = 21,799 patients with paired 12-lead ECG recordings and demographic characteristics), we trained a two-stage fusion model that takes an ICA-based noise reduction network and a wavelet decomposition network and CNN-LSTM network topped with an ensemble-based demographic analysis. The data were divided into training (70%), validation (15%) and test (15%) and stratified sampled. The SMOTE oversampling was used to deal with class imbalance. The proposed framework achieved accuracy of 97.83% (95% CI: 97.12–98.54%), sensitivity of 98.12% (95% CI: 97.45–98.79%), and specificity of 97.54% (95% CI: 96.82–98.26%) on the held-out test set. Compared to the baseline XAI-HD framework, MFF-HD demonstrated improvement (McNemar's test, p < 0.001; Bonferroni-corrected). SHAP-based explainability components ensure clinical interpretability. Limitations include absence of external validation and single-center data origin.