Adaptive Hybrid Activation PPG-based Model for Cardiovascular Disease Detection and Classification

Background: Cardiovascular diseases (CVDs) represent a significant global health burden, frequently leading to compromised arterial blood supply to vital organs. Timely and accurate detection is paramount for effective clinical intervention and improved patient outcomes. While traditional deep learning methodologies have shown promise in CVD detection from physiological signals, they often face limitations in predictive accuracy, model interpretability, and generalization across diverse data distributions. Objective: This study introduces a novel Adaptive Hybrid Activation (ADHA) based Deep Convolutional Neural Network model. The primary aim is to advance the state-of-the-art in CVD detection and classification by addressing existing limitations in accuracy, enhancing model generalization, and improving the inherent interpretability of the predictive framework. Methods: The ADHA architecture is meticulously designed with an innovation: an adaptive hybrid activation function module. The adaptive hybrid activation function is engineered to dynamically optimize non-linearity and pattern learning, thereby bolstering the model's classification efficacy. Results: Empirical evaluation, conducted on a single, publicly available dataset (Challenge 2015), demonstrates that the proposed ADHA model achieves notable performance metrics, including 98.05% accuracy, 98.00% F1-score, 98.60% sensitivity, 98.20% specificity, 98.54% negative predictive value (NPV), and 97.56% positive predictive value (PPV). These results indicate superior performance compared to several established state-of-the-art CVD detection methodologies on the evaluated dataset. Conclusion: The ADHA model significantly contributes to the enhancement of CVD detection and classification by addressing challenges related to generalization, feature sparsity, and overfitting inherent in deep learning applications. The presented results underscore the model's compelling performance and suggest its potential for future clinical investigation and application in cardiovascular health monitoring.