Incremental Value of Perivascular Adipose Tissue Radiomics for Non-contrast CT-Based Detection of Aortic Dissection: a Multicenter Study.
Authors
Affiliations (7)
Affiliations (7)
- The First School of Medicine, School of Information and Engineering, Wenzhou Medical University, Wenzhou, 325000, China.
- Department of Radiology, Ruian People's Hospital, Wenzhou, 325200, China.
- Department of Radiology, The Second Affiliated Hospital and Yuying Children's Hospital of Wenzhou Medical University, Wenzhou, 325000, China.
- Department of Radiology, Wenzhou Central Hospital, Wenzhou, 325000, China.
- Department of Radiology, The First Affiliated Hospital of Wenzhou Medical University, Wenzhou, 325000, China.
- The First School of Medicine, School of Information and Engineering, Wenzhou Medical University, Wenzhou, 325000, China. [email protected].
- Department of Radiology, The First Affiliated Hospital of Wenzhou Medical University, Wenzhou, 325000, China. [email protected].
Abstract
Aortic dissection (AD) requires prompt diagnosis, but non-contrast computed tomography (NCCT) has limited sensitivity. Perivascular adipose tissue (PVAT) may reflect vascular inflammation; however, its diagnostic contribution on NCCT and incremental value beyond aortic radiomics remain unclear. This retrospective multicenter study included 581 patients from one center for model development and internal testing and 280 patients from three external cohorts. Six models (Clinical, Aorta, PVAT, Aorta-PVAT, Aorta-Clinical, and Aorta-PVAT-Clinical) were evaluated using support vector machine (SVM), extreme gradient boosting (XGBoost), and random forest (RF) classifiers. In internal testing, area under the receiver operating characteristic curve (AUC) values were 0.876-0.895 for Aorta models, 0.727-0.764 for PVAT models, and 0.873-0.895 for Aorta-PVAT models. Corresponding external AUCs were 0.898-0.921, 0.809-0.866, and 0.896-0.908, respectively. Aorta models significantly outperformed the corresponding PVAT models internally. Across classifiers, adding PVAT radiomics did not consistently improve discrimination or decision-curve net benefit over Aorta models and did not improve Aorta-Clinical models. Feature-importance analyses likewise showed stronger contributions from aortic radiomic and clinical variables than from PVAT features in combined models. Aortic radiomics provided the dominant NCCT imaging signal for AD detection and generalized well across centers. PVAT radiomics showed independent but weaker discrimination and limited incremental value beyond aortic radiomics.