Adaptive Deep Learning for Multimodal Cardiac Risk Prediction: A Feature Fused Multichannel Approach

Cardiovascular disease (CVD) continues to be a leading global cause of death and requires advanced models and personalized risk prediction algorithms. This research presents an adaptive deep learning framework that combines feature selection and signal fusion to improve individual level cardiac risk prediction. The adaptive deep learning model collects multimodal physiological signals, including Electrocardiogram (ECG), photoplethysmogram (PPG), and other bio-signals, to create an overall health profile for each patient. The feature selection component of the adaptive model enhances the model performance by reducing noise and dimensionality of the input, enhancing learning efficiency. To enhance data representativeness, a multi-signal or multi-channel fusion paradigm was applied, taking advantage of feature correlations across multiple signals to result in a more accurate and robust representation of cardiovascular health status. The proposed predictive architecture combines convolutional neural networks (CNNs) and long short-term memory (LSTM) networks to efficiently analyse both spatial and temporal dependencies present in the signals. In testing using a public, benchmark dataset containing over 10,000 patient records, the model achieved an accuracy of 94.5%, precision of 93.8%, recall of 92.3%, and an F1 score of 93.0%. The entire diagnostic system enables remote monitoring and highly accurate and predictive results in real-time. The proposed research represents the first adaptive deep learning approach to signal fusion for robust and personalized CVD risk prediction, while addressing existing challenges within predictive health care systems.