ECG-Based Characterization of Heart Diseases by leveraging AI techniques

This research analyses three distinct approaches to improve the diagnosis and classification of cardiovascular conditions through the utilisation of machine learning techniques applied to electrocardiographic (ECG) data. The first study proposes a predictive model based on a two-dimensional convolutional neural network (CNN-2D) and Gradient-weighted Class Activation Mapping (Grad-CAM) to analyse ECG images, offering an interpretable solution for detecting myocardial infarctions and arrhythmias. The second study develops a CNN to distinguish between arrhythmias and other types of disturbances, utilising data from 1,052 patients and achieving an overall accuracy of 90%, with promising results but challenges related to signal noise and device variability. Finally, the third study explores three distinct approaches, integrating CNN, heart rate variability (HRV) feature extraction, and classification models such as Random Forest; this last approach proves to be the most effective, reaching an accuracy of 92.01%, demonstrating potential for supporting clinical decision-making.