An Intelligent Deep Learning Framework for Cardiac Arrhythmia Detection and Risk Prediction
Abstract Cardiac Arrhythmia detection using continuous ECG monitoring is challenging because abnormal heartbeats occur irregularly, and their patterns may vary from one person to another. ECG signals can also contain noise caused by body movements, muscle activity, and baseline variations, which can affect the accuracy of detection. To address these challenges, this project proposes an intelligent deep learning framework that combines a 1D Convolutional Neural Network (CNN), a Bidirectional Long Short-Term Memory (BiLSTM) network, and an Attention Mechanism to detect and classify heartbeats into five AAMI classes. The CNN extracts important features from ECG signals, while the BiLSTM learns the sequential patterns of heartbeats. The CNN extracts important features from ECG signals, particularly the QRS complex, while reducing the influence of less relevant information and noise. To improve the interpretability of the results, a gradient-based Explainable AI (XAI) method is used to generate saliency heatmaps that highlight important regions of the ECG signals. The framework also includes a risk stratification module that categorizes detected cases into low, medium, and high-risk levels based on predefined criteria. The proposed model is evaluated using the MIT-BIH Arrhythmia Database, with ECG signals segmented into 0.6 second windows around the R-peaks. The target is to achieve high classification accuracy while providing understandable results and useful risk information. This framework aims to support continuous ECG monitoring and assist healthcare professionals in identifying abnormal heartbeat more efficiently.
Authors
- Nagaraj S
- Gurusri A.M
- Madhumitha G
- Viveka A
- Poojitha P
Institutions
- Karpagam Academy of Higher Education (IN)
Publication Details
- Journal
- Research Square
- Published
- 2026-11-14
- DOI
- https://doi.org/10.21203/rs.3.rs-11484963/v1
- Primary Topic
- ECG Monitoring and Analysis
- Type
- preprint