An automatic patient-adapted ECG heartbeat classifier allowing expert assistance

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• Electrocardiogram (ECG) analysis is crucial for diagnosing cardiac conditions.
• Automatic ECG heartbeat classification algorithms face challenges in accuracy and adaptability.
• Existing methods often require significant expert annotation, increasing workload.

Purpose of the Study:

• To develop a patient-adaptable algorithm for ECG heartbeat classification.
• To integrate automatic classification with user-assisted learning for improved performance.
• To evaluate the algorithm's effectiveness across multiple ECG databases.

Main Methods:

• Developed a hybrid algorithm combining an automatic classifier and a clustering algorithm.
• Utilized features from RR interval series and wavelet transform-based morphology descriptors.
• Integrated decisions from both algorithms for automatic and assisted classification modes.
• Evaluated performance using standard ECG databases and metrics (Accuracy, Sensitivity, Positive Predictive Value).

Main Results:

• The patient-adaptable algorithm demonstrated improved performance over the original automatic classifier.
• Even with minimal manual annotations (less than 2 per recording), significant gains in accuracy (6.9%), sensitivity (6.5%), and positive predictive value (8.9%) were observed.
• Assisted mode with 12 manual annotations per recording yielded substantial improvements: 13.1% in accuracy, 13.9% in sensitivity, and 36.1% in positive predictive value.
• The assisted mode outperformed state-of-the-art classifiers with reduced expert effort.

Conclusions:

• The proposed patient-adaptable algorithm enhances ECG heartbeat classification accuracy and efficiency.
• Effective handling of expert assistance significantly improves automatic classifier performance.
• This approach offers a promising solution for more accurate and adaptable automated ECG analysis.