An adapting system for heartbeat classification minimising user input.

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Electrocardiogram (ECG) analysis is crucial for diagnosing cardiac conditions.
• Accurate heartbeat classification is essential for reliable ECG interpretation.
• Current methods often require extensive manual input from experts.

Purpose of the Study:

• To develop an adaptive system for ECG processing and heartbeat classification.
• To minimize the need for user input in the classification process.
• To enhance the accuracy and efficiency of automated ECG analysis.

Main Methods:

• A global classifier initially annotates heartbeats.
• Beat annotations are ranked by confidence scores derived from probability estimates.
• An expert validates and corrects a subset of low-confidence annotations.
• The system adapts by training a local classifier with corrected beats and integrating it with the global classifier.

Main Results:

• The adaptive system demonstrated a significant improvement in classification performance.
• Adaptation using a small number of expert-corrected beats yielded substantial performance boosts.
• The system effectively reduced the reliance on extensive manual annotation.

Conclusions:

• Adaptive learning significantly enhances automated ECG heartbeat classification.
• Minimal expert intervention can lead to substantial improvements in system accuracy.
• This approach offers a more efficient and accurate method for ECG analysis.