Method for Solving Difficulties in Rhythm Classification Caused by Few Samples and Similar Characteristics in...

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Accurate electrocardiogram (ECG) analysis is crucial for diagnosing heart disease.
• Classifying cardiac rhythms is challenging due to limited data samples and similar rhythm characteristics.

Purpose of the Study:

• To develop a novel method for accurate ECG rhythm classification, addressing limitations of small sample sizes and rhythm similarity.
• To improve the differentiation of challenging rhythms such as atrial fibrillation (AFIB) and atrial flutter (AFL).

Main Methods:

• Proposed a novel method utilizing beat score map (BSM) images derived from ECG signals.
• Incorporated associations between beats and traditional features like the R-R interval.
• Employed a convolutional neural network (CNN) model trained with transfer learning on BSM images for rhythm classification.

Main Results:

• The proposed method demonstrated significant performance improvements for ECG rhythms with small data samples.
• Achieved superior differentiation between AFIB and AFL rhythms compared to existing methods.
• Showcased a 20% performance increase for rhythms with few samples and a 30% F-1 score improvement for AFIB/AFL classification.

Conclusions:

• The novel BSM image-based method effectively overcomes limitations associated with small sample sizes and similar ECG rhythms.
• This approach offers a promising advancement in automated ECG analysis and cardiac arrhythmia diagnosis.