Deep learning based coronary cta automatic segmentation method
By using a deep learning framework based on nnU-Net, automatic segmentation of coronary CTA images was achieved, solving the problems of high workload and inaccurate segmentation caused by manual image reading. This improved the accuracy and consistency of segmentation, reduced the complexity of model training and deployment, and met the needs of rapid clinical application.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- THE FIRST AFFILIATED HOSPITAL OF ANHUI MEDICAL UNIV
- Filing Date
- 2026-04-22
- Publication Date
- 2026-07-17
AI Technical Summary
In existing coronary CTA technology, manual image segmentation and identification are labor-intensive and experience-dependent. Furthermore, existing methods are difficult to accurately segment vascular branches and small vessels, resulting in poor consistency and accuracy of results. Deep learning models are complex to train and difficult to deploy.
An automatic coronary artery CTA segmentation method based on the nnU-Net adaptive deep learning framework is adopted, including data preprocessing, model building and training, prediction and post-processing. The segmentation results are optimized by using the 3DU-Net architecture, a hybrid loss function and five-fold cross-validation, combined with morphological operations.
It enables automatic segmentation of coronary CTA images, reducing the workload of doctors, improving the accuracy and consistency of segmentation, reducing reliance on experience, enhancing the model's generalization ability and deployment efficiency, and meeting the needs of rapid clinical application.
Smart Images
Figure CN122415646A_ABST