Every year, around 3,000 abdominal aortic aneurysms in the Netherlands are treated with EndoVascular Aneurysm Repair (EVAR). The procedure is safer up front than open surgery, but 1 in 5 patients needs a re-intervention within five years. Every patient faces lifelong annual imaging, and for most of them it turns out to be unnecessary.
As a postdoctoral researcher on the project, I design, implement and train its two prediction models. They are multi-modal models that combine structured clinical data with CTA imaging, trained on the RADAR database of over 1,000 EVAR patients. The models also embed explainable AI (XAI) modules to provide clinicians with full transparency into how clinical and imaging inputs drive each prediction.
Both models will be tested in a prospective clinical study with around 300 patients across three hospitals. Their outputs feed the project's shared decision tools, from 3D prints and holograms to clinical software, so surgeons and patients can decide on treatment and follow-up together.
Multi-modal AI
Risk prediction
Clinical data + CTA fusion
Explainable AI (XAI)
Prospective validation
Model 01 · Preoperative
Optimizing the treatment plan
Predicts the 5-year re-intervention risk of an intended operative plan from preoperative imaging and clinical data. Surgeons can then adjust the stent graft choice or add adjunctive measures before the first incision.
Model 02 · Postoperative
Personalizing follow-up
Extends the first model with procedural and 30-day postoperative data to stratify follow-up: strict surveillance for high-risk patients, far fewer hospital visits for the 60–80% who remain complication-free.
Impact
Less burden, better decisions
Fewer unnecessary scans and visits, lower costs, and predictions patients can actually understand. The tools that carry these predictions are tested with the Harteraad patient panel.
