A multi-label classification system for anomaly classification in electrocardiogram

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Automatic classification of electrocardiogram (ECG) signals is a significant research area.
• Current methods primarily focus on single-label classification, which is insufficient for ECG segments potentially indicating multiple cardiac diseases.
• Single-label classification on segmented beats can disregard crucial contextual information within the ECG signal.

Purpose of the Study:

• To address the limitations of single-label ECG classification.
• To develop a more accurate method for identifying multiple cardiac diseases from ECG signals simultaneously.
• To leverage deep sequence models for enhanced ECG signal analysis.

Main Methods:

• Proposed a multi-label classification approach using the binary correlation transformation method.
• Constructed a deep sequence model to classify ECG signals.
• Transformed the multi-label problem into multiple binary classification tasks to simplify learning.

Main Results:

• Achieved an F1 score of 0.767.
• Recorded a Hamming Loss of 0.073.
• Obtained a Coverage score of 3.4 and a Ranking Loss of 0.262.
• Demonstrated superior performance compared to existing methodologies.

Conclusions:

• The proposed multi-label ECG classification method effectively identifies multiple cardiac diseases.
• Binary correlation combined with deep sequence models offers a promising direction for ECG signal analysis.
• This approach overcomes the limitations of single-label classification and preserves signal context.