A framework for comparative study of databases and computational methods for arrhythmia detection from single-lead...

Area of Science:

• Computational cardiovascular diagnostics
• Artificial intelligence in medicine
• Biomedical signal processing

Background:

• Arrhythmia detection from electrocardiogram (ECG) is crucial but hampered by limited dataset usage and inconsistent evaluation methods.
• Lack of generalizability and real-world applicability assessment for existing ECG classification techniques is a significant challenge.
• The rise of wireless single-lead ECG monitoring necessitates adaptable and reliable arrhythmia detection solutions.

Purpose of the Study:

• To introduce ECGDL, an open-source, configurable codebase for ECG classification, addressing limitations in current arrhythmia detection research.
• To provide a comprehensive comparative analysis of different deep learning models, segmentation techniques, and evaluation schemes for arrhythmia classification.
• To facilitate the assessment of ECG classification method generalizability and real-world performance, particularly for single-lead ECG data.

Main Methods:

• Developed ECGDL, an open-source framework integrating 9 arrhythmia datasets and 4 deep learning architectures.
• Implemented 4 signal segmentation techniques and 4 evaluation schemes within the ECGDL framework.
• Unified dataset class information using a label dictionary and proposed an inter-patient cross-validation scheme for fair evaluation.

Main Results:

• ECGDL enables comparative analysis across diverse ECG datasets, deep learning models, and segmentation methods.
• The framework facilitates a comprehensive understanding of arrhythmia classification performance.
• The proposed inter-patient cross-validation offers a more rigorous evaluation of model generalizability.

Conclusions:

• ECGDL provides a flexible and unified platform for advancing arrhythmia detection research using deep learning.
• The codebase aids in evaluating the generalizability and real-world performance of ECG classification methods.
• This work contributes to more reliable and standardized computational ECG analysis, especially for single-lead monitoring.