DIAGNOSTIC SUPPORT SYSTEM BASED ON AUTOMATIC ELECTROCARDIOGRAM (ECG) ANALYSIS

Abstract

Cardiovascular diseases are among the leading causes of death worldwide, making it essential to develop technologies that support the early identification of cardiac abnormalities. This work proposes a decision-support system based on the automatic analysis of electrocardiogram (ECG) signals, using one-dimensional convolutional neural networks implemented in Python with PyTorch. Two architectures were evaluated a simple CNN and a 1D ResNet , the latter being adopted for its better performance. The public PTB-XL dataset was used, with 12-lead records at 100 Hz grouped into five diagnostic superclasses (NORM, MI, STTC, CD and HYP). The train/validation/test split followed the official patient-level stratification, avoiding data leakage, and class imbalance was handled by loss weighting. The system reached 69.6% accuracy and a 0.89 macro-AUC on the test set. The goal is to support medical analysis without replacing professional diagnosis.