Detection and Classification of Chronic Total Occlusion lesions using Deep Learning

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Cardiovascular disease (CVD) poses a significant mortality risk.
• Chronic total occlusion (CTO) is a key factor affecting percutaneous coronary intervention (PCI) success rates for CVD.
• Accurate detection and classification of CTO are crucial for effective treatment planning.

Purpose of the Study:

• To develop novel deep convolutional neural networks (CNNs) for automated CTO detection and morphology classification.
• To enhance the accuracy of identifying CTO entry points in coronary angiography.
• To improve the classification of CTO morphology for better procedural outcomes.

Main Methods:

• Utilized deep convolutional neural networks (CNNs) with feature pyramid networks (FPN) for CTO detection.
• Employed a model fusion technique within the detection network.
• Implemented data augmentation and attentive regularization loss with a reciprocative learning algorithm for CTO classification.
• Trained and validated the models on a dataset of 2059 coronary angiograms annotated by cardiologists.

Main Results:

• Achieved a recall of 89.3% for CTO entry point detection.
• Reached a sensitivity of 94.5% and specificity of 89.1% for CTO morphology classification.
• Demonstrated the efficacy of the proposed CNN models in analyzing coronary angiograms.

Conclusions:

• The developed deep CNNs provide an effective automated solution for CTO detection and classification.
• These AI-driven tools can significantly aid cardiologists in diagnosing and planning PCI for CVD patients.
• The study highlights the potential of advanced AI in improving cardiovascular intervention outcomes.