A lightweight and robust method for electrocardiogram anomaly detection and localization using multi-scale masked...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Electrocardiogram (ECG) analysis is vital for diagnosing cardiovascular diseases.
• Traditional methods often require extensive labeled data and complex preprocessing like R-peak detection.
• Anomaly detection offers a flexible approach for rare and diverse cardiac conditions.

Purpose of the Study:

• To develop an efficient and robust ECG anomaly detection method.
• To overcome the limitations of existing methods relying on complex preprocessing.
• To capture both global and local dependencies in ECG signals effectively.

Main Methods:

• Proposed MMAE-ECG, a multi-scale masked autoencoder.
• Integrated multi-scale masking and attention mechanisms with distinct positional embeddings.
• Utilized a lightweight Transformer encoder and an aggregation strategy for anomaly scoring.

Main Results:

• Achieved state-of-the-art performance in ECG anomaly detection and localization.
• Significantly reduced computational costs: ~1/78 inference FLOPs and 1/18 trainable parameters.
• Demonstrated the effectiveness of multi-scale strategies and aggregation for anomaly detection.

Conclusions:

• MMAE-ECG provides an effective and efficient solution for ECG anomaly detection.
• The multi-scale masked autoencoder approach eliminates the need for complex preprocessing steps.
• This method holds significant potential for improving cardiovascular diagnostics through automated ECG analysis.