Myocardial Infarction Identification in Electrocardiogram Signals Using Long Short Term Memory Transfer Learning (LSTM-TL)

Myocardial Infarction (MI) continues to be a major cause of death globally, emphasizing the need for rapid and accurate detection methods. Electrocardiogram (ECG) signals are commonly used for cardiac diagnosis, but the complexity and variety of ECG patterns make manual interpretation difficult. This study investigates the use of a Long Short-Term Memory (LSTM) network combined with Transfer Learning (TL) for the automatic detection of MI. Using the PTB-XL dataset we implemented a single LSTM model, the most basic model of LSTM. We also performed hyperparameter tuning to further optimize the model. The results showed that TL significantly improved accuracy from 76.70% to 89.36%, with a performance increase of up to 13%. This study demonstrates the potential of TL in ECG analysis, though it is currently limited by the use of a single LSTM model and dataset. Future studies should broaden the scope by using more diverse datasets and more complex alternative model architectures to assess the generalizability of this approach in medical applications.