This analysis demonstrates machine learning enhances segmentation accuracy of plaque features in OCT, indicating better cardiovascular risk assessment.
Key Points
Automated segmentation achieved high accuracy for lumen (DSC: 0.987) in a diverse cohort of 103 patients.
The study utilized advanced machine learning models, including U-Net and DeepLabV3, optimizing accuracy for complex plaque structures.
A hybrid segmentation strategy employed single-class models for common features and multi-class models for complex morphologies.
Integration of models into a weighted ensemble significantly improved overall segmentation accuracy to a DSC of 0.882.