Optimal Multi-Stage Arrhythmia Classification Approach.

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Arrhythmia, a heart rate or rhythm problem, necessitates early diagnosis for effective treatment.
• Current diagnostic methods for arrhythmia can be improved for greater accuracy and efficiency.

Purpose of the Study:

• To develop and optimize a multi-stage arrhythmia classification scheme using 12-lead ECGs.
• To achieve accuracy comparable to professional cardiologists in identifying arrhythmia types.

Main Methods:

• A three-step noise reduction process involving Low Band Pass filter, Robust LOESS, and Non Local Means smoothing.
• A novel feature extraction method based on percentiles of interval lengths and peak/valley magnitudes.
• An optimal classification model using Extreme Gradient Boosting Tree with hyperparameter tuning.

Main Results:

• An F1-Score of 0.988 was achieved on patients without additional cardiac conditions.
• An F1-Score of 0.97 was achieved on patients with additional cardiac conditions.
• An average 10-fold cross-validation F1-Score of 0.992 was obtained on the MIT-BIH arrhythmia database.

Conclusions:

• The optimized multi-stage approach provides cardiologist-level accuracy for automatic ECG data analysis.
• The method demonstrates robust compatibility with various ECG data sources.
• This advancement can significantly improve the early detection and management of cardiac arrhythmias.