Transfer Learning Strategies for Cardiovascular Disease Detection in ECG Imagery.

Area of Science:

• Artificial Intelligence in Medical Imaging
• Deep Learning for Cardiovascular Disease Detection
• Computational Pathology

Background:

• Coronary artery disease (CAD) is a leading global cause of death.
• Traditional manual scoring of coronary artery calcium is subjective and time-consuming.
• AI offers a more efficient and accurate alternative for CAD assessment.

Purpose of the Study:

• To develop a deep learning model for automated coronary artery calcium scoring.
• To enhance accuracy and reduce bias in automated scoring systems.
• To improve clinical decision-making for early CAD diagnosis.

Main Methods:

• Development of a novel deep learning architecture, HeProbAtt BiGRU Net.
• Utilized a dataset of 14,127 non-contrast computed tomography (NCCT) slices.
• Employed classification (healthy vs. non-healthy) and regression tasks on NCCT data.

Main Results:

• The HeProbAtt BiGRU Net achieved 99% accuracy, 99% F1-score, and 0.99 ROC-AUC for classification.
• Achieved Mean Absolute Error (MAE) of 0.065 and Root Mean Squared Error (RMSE) of 0.145 for regression.
• Attention and probabilistic weights improved learning efficiency and decision precision.

Conclusions:

• The HeProbAtt BiGRU Net offers a highly accurate and efficient automated method for coronary artery calcium scoring.
• The hybrid framework supports real-time classification and regression for early CAD diagnosis.
• Future research should involve multi-center validation and explain-ability features.