Alibaba DAMO Academy and FAHZU Deploy iAorta AI for Acute Aortic Syndrome Detection in 10 Hospitals

Title

Alibaba DAMO Academy and FAHZU Deploy iAorta AI for Acute Aortic Syndrome Detection in 10 Hospitals

Content

iAorta, an AI system co-developed by Alibaba DAMO Academy and The First Affiliated Hospital of Zhejiang University (FAHZU), reduced the misdiagnosis rate of acute aortic syndrome (AAS) from 48.8% to 4.8% — a 90% reduction — and shortened average time-to-diagnosis from 4.3 hours to 1.7 hours across a pilot deployment now running in 10 hospitals in Zhejiang Province, China. According to Dr. Zhang Hongkun, Director of the Vascular Surgery Department at FAHZU: 'We are thrilled to work with Alibaba to promote this AI technology, as it can greatly benefit hospitals where medical resources and physician expertise are relatively limited. Our goal is to improve the overall diagnostic and treatment capabilities for aortic diseases across all healthcare institutions.' The clinical motivation is stark: untreated AAS carries 40–50% mortality within 48 hours, rising 1–2% per hour of delay. Although CT angiography (CTA) is the gold standard for AAS, more than half of Chinese ED patients with acute chest pain initially receive non-contrast CT due to cost and workflow constraints, and CTA is reserved for higher-suspicion cases. Non-contrast CT alone has historically lacked adequate sensitivity and specificity for AAS detection. iAorta bridges this gap by extracting reliable AAS warnings directly from non-contrast scans already being acquired. The system was trained on 3,350 aortic CTA scans with paired arterial and non-contrast phase series, using image registration to transfer arterial-phase labels to the non-contrast phase. Validation proceeded in four stages: Stage I multi-center retrospective study (n = 20,750, AUC 0.958, 95% CI 0.950–0.967); Stage II reader study (n = 2,287); Stage III large-scale real-world study across eight hospitals (n = 137,525, sensitivity 0.913–0.942, specificity 0.991–0.993); Stage IV prospective deployment including a comparative study at FAHZU (n = 13,846, time-to-correct-diagnostic-pathway 219.7 → 61.6 minutes) and a pilot at Shanghai Changhai Hospital from 20 December 2024 to 28 February 2025 (n = 15,584, 21 of 22 AAS cases correctly identified, sensitivity 0.955, specificity 0.994). iAorta is a deep-learning system that outputs (a) patient-level AAS probability, (b) segmentation masks of the aortic wall and true lumen, and (c) activation maps highlighting lesion regions. Integrated into hospital PACS, it triggers popup alerts in the diagnostic radiology interface for positive cases. The work was first posted as an arXiv preprint in June 2024 and published in Nature Medicine on 20 August 2025. DAMO iAorta entered China's NMPA Innovation Medical Device Special Review Procedure in the 2026 #8 batch, accelerating its path to commercial clearance. In the Changhai pilot, the average diagnostic time for the 21 flagged AAS patients was 102.1 minutes (range 75–133), versus the historical baseline of 4.3 hours. The team positions iAorta as a blueprint for screening other time-sensitive conditions — non-ST elevation acute coronary syndrome, pulmonary embolism, and esophageal rupture — toward a 'one-scan, multiple-screening' ecosystem. Christoph A. Nienaber, lead of the International Registry of Acute Aortic Dissection (IRAD), described the system as 'a new transformation of non-contrast CT in the emergency setting.'