Dynamic attention aggregation network for automatic segmentation of anomalous left coronary artery origin from pulmonary artery in CT angiography
10.13929/j.issn.1003-3289.2025.02.029
- VernacularTitle:基于动态注意力聚合网络自动分割CT血管造影中异常起源于肺动脉的左冠状动脉
- Author:
Rongshuo ZHENG
1
;
An ZENG
;
Jingliang ZHAO
;
Dan PAN
;
Xiaowei XU
Author Information
1. 广东工业大学计算机学院,广东 广州 510006
- Publication Type:Journal Article
- Keywords:
coronary vessels;
tomography,X-ray computed;
artificial intelligence
- From:
Chinese Journal of Medical Imaging Technology
2025;41(2):320-325
- CountryChina
- Language:Chinese
-
Abstract:
Objective To observe the value of dynamic attention aggregation network(DAANet)for automatic segmentation of anomalous origin of the left coronary artery from the pulmonary artery(ALCAPA)in CT angiography(CTA).Methods CTA data in 30 patients with ALCAPA syndrome were retrospectively enrolled.ALCAPA were segmented in CTA with DAANet based on residual edge feature(REF),pyramid dynamic attention(PDA)and dynamic global feature aggregation(DGFA)modules.Taken physician manual segmentation results as references,the results were compared with those of other networks.Results Compared with other networks,DAANet had better value for automatic segmentation of ALCAPA in CTA,which could show the details of irregular margins of left coronary artery with good continuity and high clarity,and the performance parameters were overall better.Conclusion DAANet had better performance for automatic segmentation of ALCAPA in CTA.
