MFB-CBRNN: A Hybrid Network for MI Detection Using 12-Lead ECGs

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering
• Machine Learning for Healthcare

Background:

• Myocardial infarction (MI) detection from electrocardiograms (ECGs) is crucial for timely treatment.
• Existing methods may struggle with the complex, multi-lead nature of ECG data.
• Deep learning offers potential for improved automated analysis of ECG signals.

Purpose of the Study:

• To propose and evaluate a novel hybrid deep learning network, the multiple-feature-branch convolutional bidirectional recurrent neural network (MFB-CBRNN), for MI detection.
• To leverage the strengths of convolutional neural networks (CNNs) and recurrent neural networks (RNNs) for comprehensive ECG feature extraction and aggregation.
• To introduce a new optimization technique, lead random mask (LRM), to enhance model robustness and accuracy.

Main Methods:

• Developed a hybrid MFB-CBRNN model integrating CNNs for lead-specific feature extraction and bidirectional LSTMs for feature aggregation.
• Implemented a novel lead random mask (LRM) optimization strategy to prevent overfitting and act as an implicit ensemble.
• Validated the model using the Physikalisch-Technische Bundesanstalt (PTB) diagnostic database with 148 MI and 52 healthy subjects via class-based and subject-based cross-validation.

Main Results:

• The MFB-CBRNN achieved a high overall accuracy of 99.90% in class-based fivefold cross-validation.
• Subject-based fivefold cross-validation yielded an overall accuracy of 93.08%, demonstrating strong generalization.
• The model performance was comparable or superior to existing state-of-the-art methods for MI detection.

Conclusions:

• The proposed MFB-CBRNN effectively detects myocardial infarction from 12-lead ECGs, showcasing significant potential for clinical application.
• The hybrid architecture and LRM optimization contribute to high accuracy and generalization capacity.
• This model can assist in real-world MI diagnostics, potentially reducing the workload for cardiologists.