Prognostic value of a non-contrast CT-based radiomics model in patients with Type-B aortic dissection after thoracic endovascular aortic repair.
Authors
Affiliations (6)
Affiliations (6)
- Jiajun Zou Department of Radiology, The First Hospital of Jiaxing, Affiliated Hospital of Jiaxing University, Jiaxing, Zhejiang Province 314000, P.R. China.
- Xinyi Shi Department of Radiology, The First Hospital of Jiaxing, Affiliated Hospital of Jiaxing University, Jiaxing, Zhejiang Province 314000, P.R. China.
- Jianqi Ni Department of Vascular Surgery, The First Hospital of Jiaxing, Affiliated Hospital of Jiaxing University, Jiaxing, Zhejiang Province 314000, P.R. China.
- Qin Jin Department of Vascular Surgery, The First Hospital of Jiaxing, Affiliated Hospital of Jiaxing University, Jiaxing, Zhejiang Province 314000, P.R. China.
- Haofeng Ding, Department of Basic Medical, Chengnan Community Healh Service center, Affiliated to Jiaxing Medical University, Jiaxing, Zhejiang Province 314000, P.R. China.
- Yifeng Shen Department of Vascular Surgery, The First Hospital of Jiaxing, Affiliated Hospital of Jiaxing University, Jiaxing, Zhejiang Province 314000, P.R. China.
Abstract
To explore the value of a non-contrast computed tomography (CT)-based radiomics model in predicting adverse prognosis in patients with type B aortic dissection (TBAD) after thoracic endovascular aortic repair (TEVAR). A retrospective analysis included data from 107 patients with TBAD undergoing TEVAR at the First Hospital of Jiaxing from January 2010 to May 2024. Patients were randomly divided into a training (n=86) and a validation cohort (n=21) at an 8:2 ratio. Radiomic features were extracted from non-contrast CT images to build a radiomic model. Univariate and multivariate logistic regression analyses identified independent clinical risk factors for constructing a clinical model. Lasso regression was used for data dimensionality reduction and model building. A clinical-radiomic model was established by integrating clinical risk factors with radiomic features. The discriminative ability, calibration, and clinical utility of the three models were evaluated using ROC curves (specificity and sensitivity), calibration curves, and clinical decision curve analysis (DCA), respectively. Ten radiomic features were selected in the training cohort. Clinical, radiomic, and clinical-radiomic models were generated using eight machine learning algorithms and exhibited good predictive performance. The clinical-radiomic model had the highest AUC values (0.943 in the training cohort, 0.875 in the validation cohort) compared to the clinical and radiomic models. Calibration curves and the Hosmer-Lemeshow test confirmed good calibration of all three models in both sets. DCA revealed that the nomogram model provided favorable clinical net benefits within a certain threshold range. A non-contrast CT-based radiomics machine learning model has high value for predicting adverse prognosis after TEVAR.