Multi-window temporal analysis for enhanced arrhythmia classification: leveraging long-range dependencies in...

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Electrocardiogram (ECG) analysis for arrhythmia classification faces challenges with high false positive rates and poor generalization across datasets.
• Conventional deep learning models often analyze short, isolated ECG segments (30 seconds), missing diagnostic features of arrhythmias like atrial fibrillation (AF) that manifest over longer durations.

Purpose of the Study:

• To introduce S4ECG, a novel deep learning architecture utilizing structured state-space models (S4), designed to capture long-range temporal dependencies in ECG data.
• To evaluate the efficacy of analyzing multiple consecutive ECG windows (up to 20 minutes) for improved multi-class arrhythmia classification and cross-dataset generalization.

Main Methods:

• Developed S4ECG, a deep learning architecture based on structured state-space models (S4).
• Jointly analyzed multiple consecutive ECG windows, extending analysis duration up to 20 minutes.
• Evaluated S4ECG on four public ECG databases for multi-class arrhythmia classification, including systematic cross-dataset evaluations for robustness.

Main Results:

• Multi-window analysis consistently outperformed single-window methods, increasing the area under the receiver operating characteristic curve (AUROC) by 1.0-11.6%.
• For AF detection, specificity improved from 0.718-0.979 to 0.967-0.998 at a fixed sensitivity, reducing false positives by 3-10 fold.
• S4 architecture demonstrated superior performance compared to convolutional neural network baselines.

Conclusions:

• The S4 architecture and multi-window analysis significantly enhance arrhythmia classification accuracy and cross-dataset generalization.
• Optimal diagnostic windows were identified as 10-20 minutes, suggesting these reflect underlying physiological timescales of arrhythmogenic dynamics.
• Findings offer practical guidance for ECG monitoring system design, improving detection of arrhythmias like AF.