Abnormal classification of electrocardiogram scatter plot based on token selection vision transformer.

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• Automated classification of electrocardiogram (ECG) scatter plots is crucial for diagnosing cardiac arrhythmias.
• Existing Convolutional Neural Network (CNN) models struggle with capturing non-local dependencies and global features due to limited receptive fields.

Purpose of the Study:

• To develop and validate a novel deep learning model for ECG scatter plot classification.
• To overcome the limitations of CNNs by effectively learning long-range dependencies and discriminative regions.
• To improve the accuracy of automated arrhythmia classification.

Main Methods:

• Proposed a Vision Transformer (ViT) network incorporating a Token Selection module.
• Segmented ECG scatter plots into patches and used Transformer encoder's self-attention for global context.
• Implemented dynamic filtering of discriminative patches (Tokens) for focused classification.

Main Results:

• Validated the model on real-world ECG scatter plot datasets.
• Achieved superior classification accuracy compared to traditional CNN-based models.
• Demonstrated effective capture of both global and key local features.

Conclusions:

• Introduced an effective and precise Vision Transformer model with token selection for ECG scatter plot classification.
• Overcame limitations of CNNs, offering a novel approach for automated arrhythmia diagnosis.
• The model shows significant potential for advancing cardiac arrhythmia diagnostic technology.