Background Aortic dissection (AD) is a life-threatening cardiovascular emergency. Preventive strategies with proven efficacy remain limited, and early identification and timely diagnostic workup are critical. This study aimed to develop and externally validate a fusion model integrating carotid ultrasound radiomic features with clinical and ultrasound semantic variables to discriminate AD from non-AD participants in a retrospective case–control setting. Methods We retrospectively enrolled 209 participants and randomly allocated them to training and test cohorts in a 7:3 ratio. Additionally, we selected 39 external participants as the validation set. We developed three diagnostic models: a carotid artery ultrasound radiomic model, a semantic model, and a fusion model combining both approaches. Subsequently, we compared the diagnostic efficacy of the three models. Results The area under the curve (AUC) metrics for the semantic, radiomic, and fusion models demonstrated values of 0.73, 0.84, and 0.94, respectively, in the training set, with corresponding values of 0.73, 0.87, and 0.93 in the test set. In the external validation set, the AUCs of the three models were 0.71, 0.81, and 0.91, respectively. Statistical analysis revealed significant differences in discriminative performance between the fusion model and other models ( p 0.05). Furthermore, the fusion model exhibited superior performance across multiple evaluation metrics, including accuracy, sensitivity, F1 score, calibration metrics, and clinical decision curve analysis. Conclusion Carotid ultrasound radiomics and semantic features showed diagnostic value for distinguishing AD from non-AD participants. The fusion model further improved discriminative performance, supporting future prospective multicenter studies to evaluate clinical utility in real-world triage workflows.
Cui et al. (Wed,) studied this question.