Heartbeats Classification Using Hybrid Time-Frequency Analysis and Transfer Learning Based on ResNet.

Area of Science:

• Cardiology and Biomedical Engineering
• Machine Learning and Artificial Intelligence
• Signal Processing

Background:

• Cardiac arrhythmia analysis relies heavily on accurate heartbeat classification.
• Traditional machine learning methods often require manual feature extraction, which can be complex and time-consuming.
• Existing methods may not fully capture the intricate morphological characteristics of electrocardiogram (ECG) signals.

Purpose of the Study:

• To propose a novel, automated heartbeat classification method for cardiac arrhythmia analysis.
• To leverage hybrid time-frequency analysis and deep learning (ResNet-101) for enhanced ECG signal interpretation.
• To improve the accuracy and efficiency of classifying various heartbeat categories.

Main Methods:

• A hybrid time-frequency analysis combining the Hilbert transform (HT) and Wigner-Ville distribution (WVD) was employed.
• 1-D ECG recordings were converted into 2-D time-frequency diagrams.
• A transfer learning classifier based on the ResNet-101 architecture was utilized for classification tasks.

Main Results:

• For 5 heartbeat categories (N, V, S, Q, F), the proposed method achieved an overall F1-score of 0.9595.
• For the 14 beat kinds in the MIT/BIH arrhythmia database, average accuracy reached 99.75% and an F1-score of 0.9016.
• The method demonstrated superior accuracy compared to existing approaches in heartbeat classification.

Conclusions:

• The proposed hybrid time-frequency analysis and ResNet-101 transfer learning method offers an effective and automated solution for heartbeat classification.
• This approach overcomes limitations of manual feature extraction by utilizing rich information from 2-D time-frequency diagrams.
• The high accuracy achieved in classifying both standard and detailed heartbeat categories highlights its potential for clinical applications in cardiac arrhythmia analysis.