Non-contrast CT-based radiomic signature for screening thoracic aortic dissections: a multicenter study

• Published in: European radiology Vol. 31; no. 9; pp. 7067 - 7076
• Main Authors: Guo, Yifan, Chen, Xiaojun, Lin, Xianda, Chen, Litian, Shu, Jiner, Pang, Peipei, Cheng, Jianmin, Xu, Maosheng, Sun, Zhichao
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.09.2021; Springer Nature B.V
• Subjects: Algorithms; Angiography; Aorta; Aortic dissection; Computed tomography; Confidence intervals; Diagnostic Radiology; Feature extraction; Health care facilities; Image contrast; Imaging; Imaging Informatics and Artificial Intelligence; Internal Medicine; Interventional Radiology; Magnetic resonance imaging; Medical imaging; Medicine; Medicine & Public Health; Neuroradiology; Performance prediction; Radiology; Radiomics; Sensitivity; Statistical analysis; Test sets; Thorax; Training; Ultrasound; Screening; Aortic dissection; Tomography, X-ray computed; Radiomics
• ISSN: 0938-7994; 1432-1084
• DOI: 10.1007/s00330-021-07768-2

Objective To develop a non-contrast CT-based radiomic signature to effectively screen for thoracic aortic dissections (ADs). Methods We retrospectively enrolled 378 patients who underwent non-contrast chest CT scans along with CT angiography or MRI from 4 medical centers. The training and validation sets were from 3 centers, while the external test set was from a 4th center. Radiomic features were extracted from non-contrast CT images. The radiomic signature was created on the basis of selected features by a logistic regression algorithm. The area under the curve (AUC) of the receiver operating characteristic (ROC) curve, accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were conducted to assess the predictive ability of radiomic signature. Results The radiomic signature demonstrated AUCs of 0.91 (95% confidence interval [CI], 0.86–0.95) in the training set, 0.92 (95% CI, 0.86–0.98) in the validation set, and 0.90 (95% CI, 0.82–0.98) in the external test set. The predicted diagnosis was in good agreement with the probability of thoracic AD. In the external test group, the diagnostic accuracy, sensitivity, specificity, PPV, and NPV were 90.5%, 85.7%, 91.7%, 70.6%, and 96.5%, respectively. Conclusions A radiomic signature based on non-contrast CT images can effectively predict thoracic ADs. This method may serve as a potential screening tool for thoracic ADs. Key Points • The non-contrast CT-based radiomic signature can effectively predict the thoracic aortic dissections. • This radiomic signature shows better predictive performance compared to the current clinical model. • This prediction method may be a potential tool for screening thoracic aortic dissections.