A visually interpretable detection method combines 3-D ECG with a multi-VGG neural network for myocardial infarction...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Artificial intelligence (AI) shows promise for recognizing myocardial infarction (MI) from conventional 1-D ECG signals.
• Existing AI methods struggle with 3-D ECG images, often lacking accuracy, interpretability, or inter-patient classification.
• Physician diagnostic support, like highlighting abnormal leads, is limited in current AI applications for ECG analysis.

Purpose of the Study:

• To develop an AI-driven automatic classification method for MI detection using 3-D ECG images.
• To achieve high inter-patient classification accuracy for MI diagnosis.
• To enhance diagnostic interpretability through visual heatmaps generated by Grad-CAM++.

Main Methods:

• A multi-VGG deep convolutional neural network was applied to top-view 3-D ECG images derived from standard 12-lead ECG signals.
• A multi-network approach separately classified QRS areas, ST areas, and whole heartbeats to boost performance.
• Grad-CAM++ was utilized to generate interpretable heatmaps, assisting physicians in MI diagnosis.

Main Results:

• The proposed method achieved 95.65% inter-patient accuracy and perfect inner-patient accuracy on the PTB diagnostic ECG database.
• On the PTB-XL database, the method attained 97.23% inter-patient accuracy.
• Grad-CAM++ highlighted areas consistent with medical criteria for MI, validating diagnostic relevance.

Conclusions:

• 3-D ECG images combined with AI classification offer an efficient approach for heart disease diagnosis.
• The developed AI method provides both high diagnostic accuracy and crucial visual interpretability.
• This technology can significantly aid physicians in diagnosing myocardial infarction.