Deep Learning and Optimization Approaches in Heart Disease Prediction Using Optical Electrocardiograms (ECG) and a Hybrid Convolutional Block Attention Capsule Network: A Review

Abstract

Heart disease remains one of the leading causes of mortality worldwide, necessitating accurate and early diagnosis for effective treatment and prevention. Optical electrocardiograms (ECG), derived from photoplethysmography (PPG) and wearable sensors, have emerged as a non-invasive and cost-effective alternative for continuous cardiac monitoring. This review explores recent advancements in heart disease prediction using optical ECG signals combined with deep learning and optimization techniques. Particular emphasis is placed on hybrid architectures integrating convolutional neural networks (CNN), block attention modules, and capsule networks. CNNs enable automatic feature extraction from raw signals, while attention mechanisms enhance feature selection by focusing on relevant signal segments. Capsule networks further improve classification performance by preserving spatial relationships and hierarchical features. Additionally, optimization techniques such as stochastic pooling, data augmentation, and hyperparameter tuning are analyzed for their role in improving model generalization and robustness. Literature from 2020–2023 demonstrates that hybrid CNN-attention-capsule models achieve superior performance compared to traditional machine learning approaches, with accuracy exceeding 95% in several studies. This review provides a comprehensive comparative analysis of existing methodologies, highlights current challenges such as noise sensitivity and computational complexity, and outlines future research directions for developing efficient, interpretable, and real-time heart disease prediction systems.