Perivascular Adipose Tissue Radiomics Predicts Abdominal Aortic Aneurysm Rupture: A Multicenter Study

Compared to stable abdominal aortic aneurysms (AAA), the inflammatory response in perivascular adipose tissue (PVAT) may be exacerbated prior to rupture, leading to functional and structural alterations that manifest as imaging disparities. Radiomics enables the extraction of images features, which can be integrated with machine learning(ML) to construct models for clinical decision support. This study investigated the potential of radiomic features derived from PVAT to predict AAA rupture.

A retrospective analysis was conducted using aortic Computed Tomography Angiography (CTA) images from two centers, comprising patients with either stable or ruptured AAA who had pre-rupture CTA scans. These images were allocated to a development set and an external validation set. After radiomic feature extraction, statistically significant features between the two groups were subjected to dimensionality reduction. Subsequently, ten common ML models were constructed and validated using both internal and external validation sets.

The development set comprised 37 ruptured patients and 155 non-ruptured patients. The external test set included 6 ruptured patients and 30 non-ruptured patients. A total of 107 radiomic features were extracted per patient, of which 18 exhibited statistically significant differences between groups. After dimensionality reduction, 5 representative features were selected. The constructed models achieved an average accuracy of 0.76 and an average AUC of 0.81 in the internal test set, while the external test set yielded an average accuracy of 0.73 and an average AUC of 0.77.

Significant differences exist in PVAT characteristics between ruptured and non-ruptured AAA patients, supporting the feasibility of using radiomic features for rupture prediction with reasonable accuracy.