What is the clinical evidence from this study?

Study design: Other. Population: Cardiovascular Disease and Obstructive Sleep Apnea (n=4300). Intervention: CNN-BiLSTM-Multi-Head Self-Attention (MHSA) framework vs. Random Forest, CNN, and BiLSTM models. Primary outcome: Classification accuracy.

July 20, 2026Iconic Research and Engineering Journals0 citationsOpen Access

A Self-Attention-Based Deep Learning Framework for Early Prediction of Cardiac Disease Using Sleep Apnea Signals

PRParamita Ray SPSwarna Ganga Priya Ponnada

Key Result

The proposed CNN-BiLSTM-MHSA deep learning framework achieved an accuracy of 98.75% and an F1-score of 98.57% in classifying sleep apnea-related physiological signals for cardiac risk prediction, outperforming traditional models.

Key Points

This study aims to develop a deep learning framework for early prediction of cardiac disease based on sleep apnea signals.
Utilized an end-to-end deep learning model integrating 1D-CNN, BiLSTM, and self-attention modules.
Analyzed physiological signals including ECG, HRV, SpO₂, and respiratory signals from public datasets.
Assessed performance through metrics like accuracy, precision, recall, F1-score, specificity, and ROC-AUC.
The model achieved significant improvements in prediction accuracy and interpretability.
Enhanced prediction of cardiovascular risks through informative segment identification.
Demonstrated utility as a decision-support tool for clinical interventions.

Structured PICO

Population

A dataset of 4,300 physiological signal segments from publicly available databases used to evaluate a deep learning model for sleep apnea and cardiac risk classification.

Intervention

Self-Attention-Based Deep Learning Framework integrating 1D-CNN, BiLSTM, and Multi-Head Self-Attention using ECG, HRV, SpO₂, and respiratory signals

Outcome

Early prediction of cardiac disease (performance assessed via accuracy, precision, recall, F1-score, specificity, and ROC-AUC)

A proposed deep learning framework utilizing sleep apnea-related physiological signals aims to provide an intelligent clinical decision-support tool for early cardiovascular risk assessment.

Main Result

Absolute Event Rate: 98.75% vs 93.42%

Limitations

The study relies on retrospective public datasets rather than real-time clinical monitoring.
Future work is needed to incorporate multimodal wearable sensor data and explainable AI techniques to enhance clinical applicability.

Abstract

cardiovascular disease (CVD) is one of the leading causes of mortality worldwide, and early detection is essential for reducing its clinical and economic burden. Obstructive sleep apnea (OSA), a common sleep-related breathing disorder, has been identified as a significant risk factor for various cardiovascular conditions, including hypertension, arrhythmias, heart failure, and coronary artery disease. This project proposes a novel end-to-end deep learning framework that utilizes sleep apnea-related physiological signals for the early prediction of cardiac disease. The proposed model integrates one-dimensional Convolutional Neural Networks (1D-CNN), Bidirectional Long Short-Term Memory (BiLSTM) networks, and a Multi-Head Self-Attention mechanism to automatically extract spatial and temporal features from electrocardiogram (ECG), heart rate variability (HRV), blood oxygen saturation (SpO₂), and respiratory signals. The self-attention module enhances the model by identifying the most informative signal segments associated with cardiovascular abnormalities, thereby improving prediction accuracy and interpretability. The framework is evaluated using publicly available sleep apnea and cardiac datasets, with performance assessed through accuracy, precision, recall, F1-score, specificity, and ROC-AUC. The proposed system aims to provide an intelligent clinical decision-support tool for early cardiovascular risk assessment, enabling timely intervention, continuous patient monitoring, and improved healthcare outcomes through AI-driven predictive analytics.

AIに質問

Bookmark

View Full Paper