Characterization of Heart Diseases per Single Lead Using ECG Images and CNN-2D.

Area of Science:

• Cardiology and Artificial Intelligence
• Medical Diagnostics
• Biomedical Signal Processing

Background:

• Cardiopathy is a leading global cause of mortality, necessitating early and accurate diagnosis.
• Electrocardiograms (ECGs) are a primary, cost-effective tool for initial cardiac assessment.
• Automated methods for cardiopathy identification are crucial for timely intervention.

Purpose of the Study:

• To develop and validate a customized Convolutional Neural Network (CNN) for automated heart disease detection using ECG data.
• To analyze the diagnostic contribution of different ECG electrode configurations (bands and single electrodes).
• To assess the performance and interpretability of the CNN model in classifying cardiac conditions.

Main Methods:

• A customized Convolutional Neural Network (CNN) architecture was designed for ECG image analysis.
• The model processed ECG data from four patient categories: three distinct heart conditions and healthy controls.
• Analysis included both multi-lead ECG signals and individual electrode data for feature extraction.

Main Results:

• The CNN model demonstrated high performance, with recall rates exceeding 80% for most analyzed heart diseases, reaching 100% in some cases.
• The study provided insights into the importance of specific electrode bands and individual electrodes for detecting particular pathologies.
• The developed model offers a degree of interpretability regarding the diagnostic significance of ECG components.

Conclusions:

• The customized CNN model effectively forecasts heart diseases from ECG images, offering a promising automated diagnostic tool.
• The findings highlight the potential of AI in enhancing the accuracy and efficiency of cardiopathy identification.
• This approach contributes to reducing mortality risk and improving treatment efficacy through early disease detection.