1D-CNN-LSTM Fused Hybrid Model for the Detection of Myocardial Infarction from Photoplethysmography Signals

Abstract

Myocardial Infarction (MI), referred to as a heart attack, is a critical cardiovascular condition resulting from restricted blood supply to a section of the heart, causing damage or death to heart muscle tissue. Early and precise identification of MI is essential to ensure timely intervention and enhance patient survival rates. Conventional diagnostic techniques such as blood-based biomarkers face significant challenges, including delayed detection, high noise sensitivity, and dependence on expert interpretation, potentially leading to misdiagnosis or treatment delays. To overcome these inherent limitations, this research introduces an innovative hybrid approach that integrates one-dimensional convolutional neural networks (1D CNN) and Long-short term memory networks (LSTM) for MI detection from Photoplethysmography (PPG) signals. 1D convolution captures and extracts local spatial features from sequential time-series data by detecting subtle physiological variations indicative of MI, while LSTM captures long-term temporal dependencies inherent in PPG signals, improving the system’s ability to detect intricate temporal features associated with MI. The suggested hybrid method was trained and validated on a publicly accessible Kaggle dataset, which is a time series dataset primarily focused on analyzing physiological signals captured using a PPG sensor. The proposed hybrid model attained an impressive overall accuracy of 96.31%, indicating a balanced and robust performance across both normal and MI classes. This high level of accuracy underscores the model’s effectiveness and potential applicability for real-world use. The findings highlight the model’s ability to identify MI with high reliability, showcasing its potential for low-cost and non-invasive cardiac health monitoring in clinical and remote settings.