AI-based Framework for Automated Segmentation and Surgical Decision Support in Ascending Thoracic Aortic Aneurysm

Accurate morphometric assessment of ascending thoracic aortic aneurysms (ATAA) from computed tomography angiography (CTA) is essential for individualized risk stratification and surgical planning. Existing automated methods for aortic segmentation and diameter quantification are often limited by morphological variability and lack integration with risk prediction. To address these limitations, we propose a fully automated Artificial Intelligence (AI) pipeline that integrates: (i) a deep learning model for fully automated ATAA-related segmentation on CTA, ii) an algorithm to estimate the maximum aortic diameter, iii) a radiomic analysis for quantitative imaging feature extraction and iv) a machine learning classifier that leverages radiomic features to predict elective surgical indication. In this retrospective study, CTA scans from 335 participants were analyzed; 74 patients (22.1%) underwent surgical intervention within 500 days (median 53 days; interquartile range, 20-130 days) after baseline imaging. The segmentation model achieved a Dice similarity coefficient of 0.98 (0.86-0.99), and automated diameter estimation yielded a mean absolute error of 1.50 mm (-1.00 to 3.13 mm) relative to expert annotations. Radiomic-based classification achieved an AUC of 0.86 (0.83-0.87) for predicting surgical candidacy. The proposed AI-based framework enables accurate, fully automated ATAA segmentation and diameter measurement from CTA. While further validation on larger datasets is needed to assess generalizability and clinical applicability, the findings suggest that such AI-driven tools may offer a structured approach to support individualized risk assessment and management in ATAAs.