A three-level risk stratification algorithm for arrhythmia based on CNN-LSTM palindromic structure.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Cardiovascular diseases necessitate early screening and intervention due to high mortality and recurrence rates.
• Real-time electrocardiogram (ECG) monitoring is challenging due to limited medical resources and complex ECG signal features.
• Accurate arrhythmia detection and risk stratification are crucial for effective patient management.

Purpose of the Study:

• To develop a three-level risk stratification algorithm for arrhythmias using ECG signals.
• To enhance the accuracy of arrhythmia detection and risk assessment.
• To optimize medical resource allocation through precise intervention strategies.

Main Methods:

• Proposed a three-level risk stratification scheme: Normal, Not Life-threatening, and Life-threatening.
• Developed a deep learning fusion network combining Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) with a palindromic structure.
• Utilized multi-dataset experiments and cross-validation for performance evaluation.

Main Results:

• The algorithm achieved high accuracy (99.68%) and specificity (99.65%) on 2-second ECG segments.
• 10-fold cross-validation showed 99.62% ± 0.09% accuracy on 3-second segments.
• Patient-level validation on 10-second segments yielded 96.5% accuracy, demonstrating reliability across varied segment lengths.

Conclusions:

• The proposed deep learning algorithm effectively performs three-level arrhythmia risk stratification and detection.
• The method demonstrates superior performance compared to standalone CNN, LSTM, and classical classifiers.
• The algorithm offers a reliable and practical solution for improving cardiovascular disease management and resource allocation.