A Survey of Methods and Architectures for Heart Disease Prediction Using Optical Electrocardiograms (ECG) and a Hybrid Convolutional Block Attention Capsule Network

Abstract

Cardiovascular diseases (CVDs) remain the leading cause of mortality worldwide, necessitating early and accurate diagnosis. Electrocardiogram (ECG) signals play a vital role in detecting cardiac abnormalities due to their ability to capture electrical heart activity non-invasively. Recent advancements in optical sensing techniques such as photoplethysmography (PPG) and remote photoplethysmography (rPPG) have enabled contactless ECG signal approximation, offering new opportunities in telemedicine and wearable healthcare systems.

Simultaneously, deep learning (DL) approaches have demonstrated significant improvements in ECG-based disease prediction by automating feature extraction and classification tasks. Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Capsule Networks (CapsNet) have been widely explored for accurate heart disease detection. However, limitations such as loss of spatial hierarchies and insufficient attention to relevant features persist.

To address these challenges, hybrid architectures combining convolutional block attention mechanisms with capsule networks have emerged as a promising solution. These models enhance feature representation, preserve spatial relationships, and improve classification accuracy. This survey provides a comprehensive overview of recent methods, focusing on optical ECG acquisition, deep learning architectures, and hybrid attention-based capsule networks. Comparative analysis highlights performance improvements, challenges, and future directions in developing efficient, scalable, and interpretable cardiac diagnostic systems.