Exploratory machine learning analysis to characterize angioscopic features associated with atherosclerosis-related aortic dissection: an exploratory single-center angioscopic study

Aim: Aortic dissection (AD) is a life-threatening condition. Non-obstructive general angioscopy (NOGA) enables direct visualization of spontaneously ruptured aortic plaques and injuries (SRAPIs). This study evaluated the diagnostic potential and pathophysiological relevance of SRAPIs to AD. Methods: This single-center, cross-sectional observational pilot study included 56 patients with AD and 444 control patients with coronary artery disease undergoing NOGA. Healthy volunteers could not be ethically included because NOGA requires invasive catheterization. SRAPIs were classified into 13 types, and their occurrence in the aorta and bilateral common iliac arteries was analyzed using machine learning. A classification model based on Random Forest was developed, with feature selection using Least Absolute Shrinkage and Selection Operator (LASSO) regression. SHapley Additive Explanations (SHAP) and Permutation Importance were applied to interpret feature contributions. Results: Intramural blood (IB) was confirmed as the most consistent and influential SRAPI associated with AD, while puff sign (P) and salmon-pink appearance (SP) also showed strong importance. Fissure bleeding (FB) was highly frequent but showed variable importance across analytical approaches. Network analysis demonstrated preserved structural relationships after SMOTE and distinct patterns between AD and control groups. Conclusion: Aortic injury in AD may involve subintimal injury represented by IB and SP, blood or fibrin adherence to the injured vascular wall represented by P. FB may reflect a critical but non-linear pathological state rather than a purely quantitative marker. These findings reflect a selected, stable-phase cohort and require validation in larger multicenter studies.