Stony Brook researchers develop artificial intelligence system to predict aneurysm growth and rupture, offering tailored imaging and follow-up recommendations for patients.
Stony Brook, NY, May 4, 2025 — Current medical guidelines for monitoring abdominal aortic aneurysms (AAA) rely heavily on a single factor: the diameter of the aneurysm. But patients are more than a number—and their care should be too. A new research initiative at Stony Brook University Hospital is rethinking how clinicians track and treat this serious vascular condition by developing an artificial intelligence-driven clinical decision support tool that takes into account the full complexity of each patient’s health data.
Funded by the AI Innovation Seed Grant and led by researchers from the Division of Vascular Surgery, Department of Radiology, and Department of Biomedical Informatics at Stony Brook, the team is combining advanced machine learning, computer vision, and natural language processing to predict AAA growth and rupture risk more accurately than ever before. The goal: to guide clinicians on when to bring patients back for follow-up and what type of imaging — CT scan or ultrasound — is more appropriate.
Apostolos Tassiopoulos
“Abdominal aortic aneurysm surveillance has traditionally applied a ‘one-size-fits-all’ approach,” said professor and chair Apostolos Tassiopoulos, Department of Surgery, Renaissance School of Medicine. “But the disease progresses differently in different patients, and imaging alone doesn’t tell the whole story. Our tool is being designed to make surveillance more precise and more personal.”
The project builds on a large database of more than 1,000 patients enrolled in the hospital’s AAA Surveillance Program. Researchers are curating this dataset to include both electronic medical records and over 5,000 CT imaging studies. Using natural language processing, they are extracting relevant clinical data, such as comorbidities like diabetes, hypertension, and smoking status, and pairing this information with detailed image analysis made possible by deep learning models.
One key innovation is the use of a 3D U-Net segmentation model, which enables automatic and accurate identification of features such as calcium deposits, thrombus, and vessel wall structure from CT scans, which are key to predicting aneurysm growth and rupture. The team has also introduced a vision transformer model that processes imaging data to capture subtle yet clinically important patterns beyond what traditional models offer. These imaging features are then merged with clinical variables in a custom-designed neural network that predicts rupture risk and aneurysm growth on a patient-by-patient basis.
“Integrating these multimodal data streams — text from radiology reports and structured EMR entries, as well as quantitative imaging features — allows us to understand risk in a much more nuanced way,” said Dr. Prateek Prasanna, Department of Biomedical Informatics and member of the AI Innovation Institute.
Beyond prediction, the tool is designed to provide real-time guidance to clinicians. For example, if a patient is flagged as low risk, the tool may recommend a longer interval before the next follow-up, reducing unnecessary visits or radiation exposure. If a patient’s profile suggests faster growth or higher rupture risk, it could prompt earlier imaging or intervention. It will also evaluate whether CT or ultrasound is the better option at the next appointment, a decision current guidelines often leave vague.
The team’s approach represents a shift in how AI can be applied in clinical practice — not just to analyze data, but to support real-time decision-making in a way that is interpretable, evidence-informed, and ultimately more human-centered.
The researchers plan to test the predictive model in a retrospective study first, with the long-term goal of launching a prospective clinical trial. The outcomes of this research will influence future surveillance guidelines. By integrating a broader range of patient-specific risk factors, the team hopes to contribute to more data-driven, personalized standards for AAA care.
“This collaboration across departments is exactly what modern clinical research needs,” said Dr. Tassiopoulos. “We’re using technology to solve a real problem in patient care, and that’s the kind of innovation that can truly make a difference.”
Communications Assistant
Ankita Nagpal