Semisupervised ECG Ventricular Beat Classification With Novelty Detection Based on Switching Kalman Filters

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Automatic analysis of electrocardiogram (ECG) signals is crucial for accurate diagnosis, especially with increasing remote monitoring.
• Existing model-based filtering approaches primarily focus on healthy subjects and known heartbeat morphologies.
• The need for automated ECG analysis is growing due to the rise in long-term recordings.

Purpose of the Study:

• To extend model-based ECG filtering to accurately classify normal, ventricular, and previously unencountered heartbeats.
• To introduce a switching Kalman filter for automatic mode selection and signal filtering.
• To enable novelty detection of unknown ECG morphologies using a dedicated 'X-factor' mode.

Main Methods:

• A switching Kalman filter approach was developed to model normal, ventricular, and unknown (X-factor) heartbeats.
• The filter automatically selects the most likely beat type while applying prior knowledge for signal filtering.
• The method was evaluated on the MIT-BIH arrhythmia and Incart databases for ventricular heartbeat classification.

Main Results:

• The proposed technique achieved superior F1 scores of 98.3% and 99.5% on the MIT-BIH and Incart databases, respectively.
• Only 3% of beats were classified as X-factor, predominantly those with high noise levels.
• The approach demonstrated accurate classification even with unseen morphologies and noise.

Conclusions:

• The novel switching Kalman filter provides accurate automated beat classification for ECG signals, including unknown morphologies and noise.
• This method enhances the analysis of arbitrary ECG leads, offering a robust solution for pathological signal diagnosis.
• The technique represents a significant advancement in automated ECG interpretation for clinical applications.