An end-end arrhythmia diagnosis model based on deep learning neural network with multi-scale feature extraction

Area of Science:

• Artificial Intelligence
• Cardiology
• Biomedical Engineering

Background:

• Arrhythmia diagnosis faces challenges requiring improved computational models.
• Current methods may lack efficiency in feature extraction and classification.

Purpose of the Study:

• To develop an end-to-end deep learning model for accurate arrhythmia diagnosis.
• To enhance feature extraction from heartbeat signals for better classification.

Main Methods:

• Developed an adaptive online convolutional network (AOCT) for classification.
• Implemented automatic extraction of time-domain, time-frequency-domain, and multi-scale features.
• Utilized multi-scale features to improve learning of complex signal information.

Main Results:

• The AOCT-based model demonstrated excellent parallel computing and classification capabilities.
• Model performance improved significantly with the integration of multi-scale features.
• Achieved an average accuracy of 99.72%, recall of 99.62%, and F1 score of 99.3% for four common heart diseases.

Conclusions:

• The proposed deep learning model offers a highly effective solution for arrhythmia diagnosis.
• Multi-scale feature extraction is crucial for enhancing the performance of deep learning diagnostic models.
• The AOCT model shows promise for real-world clinical application in cardiology.