Improved state refinement for LSTM determined 3D CAISR-LSTM model for automatic myocardial infarction detection.

Area of Science:

• Artificial Intelligence in Medicine
• Cardiovascular Diagnostics
• Signal Processing

Background:

• Electrocardiograms (ECGs) are crucial for diagnosing myocardial infarction (MI), but interpretation requires expert cardiologists, leading to resource limitations and diagnostic delays.
• Automating MI detection from ECGs is essential to overcome these challenges and enable timely patient care.

Purpose of the Study:

• To develop and evaluate a novel automatic detection approach for MI using 12-lead ECG data.
• To implement an improved state refinement for a Long Short-Term Memory (LSTM) determined 3D Convolution-Attention (3D CAISR-LSTM) model for enhanced MI detection accuracy.

Main Methods:

• A 3D CAISR-LSTM model was developed and trained end-to-end on preprocessed 12-lead ECG signals.
• ECG signals were transformed into time-frequency images using continuous wavelet transform and bilinear interpolation.
• The model incorporated a convolutional module, attention modules, and an improved LSTM state refinement, evaluated using ten-fold cross-validation on the PTB diagnostic ECG database.

Main Results:

• The 3D CAISR-LSTM model achieved high performance metrics: 98.45% accuracy, 98.69% sensitivity, 97.50% specificity, and 99.03% F1 score.
• The proposed model significantly outperformed existing advanced 2D and 3D deep neural network architectures in MI detection.

Conclusions:

• The 3D CAISR-LSTM model demonstrates significant potential for accurate and early automatic detection of myocardial infarction from ECGs.
• This approach could lead to the development of lightweight, embedded MI detection equipment for widespread clinical use and early warning systems.