Development and validation of an AI decision support system for automated 17-zone mapping and subtype diagnosis of acute aortic syndrome on CTA

Background Acute aortic syndrome (AAS), including aortic dissection (AD), intramural hematoma (IMH) and penetrating aortic ulcer (PAU), is a life-threatening emergency with 40-50% mortality within 48 hours without prompt diagnosis1,2. Computed tomography angiography (CTA) is the diagnostic gold standard, but manual interpretation relies on senior radiologists, causing delays in resource-limited settings3. AI-assisted AAS tools lack standardized 17-zone aortic segmentation (extended SVS/STS classification), robust multi-institutional validation, and integration with slice-level subtype classification. Purpose This study aimed to develop and externally validate an AI-powered AAS decision support system (AAS-DSS) for automated 17-zone segmentation and slice-level AD/IMH/PAU classification, and to compare its diagnostic accuracy and efficiency with radiologists for clinical translatability. Methods AAS-DSS integrated nnU-Net V2 with TotalSegmentor (17-zone segmentation: 10 trunk + 7 branches) and ResNet-18 (slice-level classification). Trained on 586 CTA scans (112 AD, 86 IMH, 113 PAU, 275 controls) and externally validated on 198 multi-institutional cases (57 AD, 27 IMH, 29 PAU, 85 controls) with distinct CT scanners (Philips vs. Siemens). A 100-point scoring system (5.882 points/zone) compared accuracy with junior (3-year) and senior (10-year, gold standard) radiologists. Non-inferiority, inter-observer agreement (Cohen’s Kappa) and post-hoc power analyses were performed, with diagnostic time quantified. Results AAS-DSS achieved excellent segmentation (mean Dice Similarity Coefficient: 0.870 internal, 0.804 external), outperforming AortaSeg24 and other SOTA frameworks. Slice-level classification accuracy: AD (0.976), IMH (0.990), PAU (0.972), with extremely low false-negative rates (≤1.77%). Externally, AAS-DSS scored significantly higher than junior radiologists (all P≤0.047) and was non-inferior to seniors (all P<0.001, margin=5.882). Diagnostic time was reduced by 66.2% vs. seniors (204.22s vs.604.01s) and 75.5% vs. juniors (204.22s vs.832.42s, all P<0.001). Post-hoc power >95% and Cohen’s Kappa>0.75 confirmed reliability. Conclusion AAS-DSS delivers high accuracy, robust cross-institutional generalizability and superior efficiency for automated 17-zone aortic mapping and AAS subtype diagnosis on CTA. Outperforming junior radiologists and non-inferior to seniors, it accelerates time-critical AAS diagnosis, standardizes lesion localization and addresses resource-limited setting gaps. Seamless PACS integration supports clinical translation.