Myocardial Infarction Detection Using Hybrid Deep Learning

Detection of Myocardial Infarction (MI) from phonocardiogram (PCG) heart sound signals using a hybrid CNN-LSTM deep learning model.

Overview

The system processes heart sound recordings and classifies them as Normal or Myocardial Infarction.

The dataset contains 560 PCG recordings from 140 subjects, collected at Hasan Sadikin Hospital, Indonesia.

Pipeline

PCG Signal → Preprocessing → Feature Extraction → CNN-LSTM → MI / Normal

Features used:

• MFCC
• Wavelet Transform
• Zero Crossing Rate
• Spectral Contrast
• RMS
• Shannon Entropy

Model

The proposed architecture combines:

• 1D CNN for feature learning
• LSTM for temporal patterns
• GridSearchCV for hyperparameter tuning