Optimizing myocardial infarction detection: a hybrid CNN-GRU deep learning approach.

Area of Science:

• Cardiovascular Medicine
• Artificial Intelligence in Healthcare
• Medical Diagnostics

Background:

• Myocardial infarction (MI) is a critical condition requiring prompt diagnosis.
• Electrocardiograms (ECGs) are vital for MI detection, but interpretation can be challenging.
• Developing advanced diagnostic support tools is crucial for improving patient outcomes.

Purpose of the Study:

• To optimize myocardial infarction (MI) detection using electrocardiograms (ECGs).
• To develop and evaluate a hybrid Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU) deep learning model (DLM) for MI diagnosis.
• To assess the DLM's performance as a diagnostic support tool in cardiovascular medicine.

Main Methods:

• A retrospective analysis of 56,354 ECGs (41,871 MI patients, 14,474 healthy controls).
• Utilized 20-second, 15-lead ECG recordings sampled at 1000 Hz, pre-processed with the Pan-Tompkins algorithm.
• Trained and validated the CNN-GRU model on 85% and 15% of the data, respectively; performance evaluated using AUC, sensitivity, and specificity.

Main Results:

• The CNN-GRU model demonstrated high diagnostic accuracy for MI detection, with an Area Under the Curve (AUC) approaching one.
• With 15 leads, the model achieved 99.43% accuracy, 99.71% sensitivity, and 98.59% specificity.
• Performance further improved using Lead II alone, reaching 99.73% accuracy, 99.75% sensitivity, and 99.66% specificity, indicating superior diagnostic capability.

Conclusions:

• The proposed CNN-GRU deep learning model shows significant potential as an effective tool for supporting clinical decision-making in MI diagnosis.
• The model's high performance, especially using Lead II, suggests its utility in enhancing cardiovascular medicine diagnostics.
• Findings support further research into AI-driven diagnostic tools for cardiovascular diseases.