Intracranial Aneurysms, Subarachnoid Hemorrhage (SAH) From Ruptured Aneurysm, Unruptured Intracranial Aneurysm
Conditions
Keywords
Artificial Intelligence, Intracranial Aneurysm, Brain Aneurysm, Unruptured Intracranial Aneurysm, MR Angiography, 3D Segmentation, Rupture/Growth Risk Prediction, Subarachnoid Hemorrhage, Clinical Decision Support, SHERPA, UMC Utrecht, Morphology, Shape
Brief summary
ANEURYSM@RISK is an observational study aiming to develop and validate an artificial intelligence (AI)-based prediction model for the growth and rupture of intracranial aneurysms (IAs). By applying automated 3D segmentation and morphological quantification of IAs from MR angiography (MRA) scans, the model is intended to provide clinicians with objective and reproducible risk estimates of aneurysm instability. The study utilizes retrospective imaging data from multiple European centers, including UMC Utrecht, AP-HP Paris, and University Medical Center Hamburg-Eppendorf (UKE). A clinical vignette study will evaluate the model's clinical utility and user experience among interventional radiologists. This study is exempt from medical ethics review (non-WMO in the Netherlands), as it involves only existing, anonymized data and imposes no additional burden on patients.
Detailed description
The ANEURYSM@RISK study is part of the SHERPA project (Smart Human-centred Effortless support for Professional clinical Applications). The study aims to develop and validate a multivariable artificial intelligence (AI)-based prediction model to identify unstable unruptured intracranial aneurysms (UIAs), using morphological and clinical features. Retrospective MR angiography (MRA) data will be collected from three clinical sites: UMC Utrecht (The Netherlands), AP-HP Paris (France), and University Medical Center Hamburg-Eppendorf (Germany). The study workflow includes: * Development of AI algorithms for 3D shape feature extraction after automated aneurysm segmentation * Training of predictive models for aneurysm growth and rupture based on morphological and clinical parameters * Validation of model performance using a longitudinal dataset of \ 1,000 patients (target C-statistic ≥ 0.80) * A clinical vignette study in real-life settings to evaluate usability, decision-making impact, and inter-clinician variability Key Performance Indicators (KPIs): * Discriminative performance for aneurysm instability prediction; C-statistic ≥ 0.80 * Sensitivity ≥ 80% and specificity ≥ 50% (based on optimal cut-off values) * ≥ 25% reduction in time to clinical decision-making * ≥ 80% adherence to AI-generated suggestions by interventional radiologists * ≥ 20% improvement in user experience using 3D visualization compared to 2D displays (survey-based) * ≥ 50% reduction in inter- and intra-observer variability in aneurysm assessment Ethical Considerations: This is a non-interventional, retrospective study using previously acquired and anonymized imaging data. No additional procedures or data collection will be performed. The study poses no added burden or risk to patients. According to Dutch regulations, it is not subject to the Medical Research Involving Human Subjects Act (non-WMO).
Interventions
None listed
Sponsors
AP-HP (Assistance Publique - Hôpitaux de Paris), FRANCE : Hôpital Pitié Salpêtrière, Hôpital Bichat
University Medical Center Hamburg-Eppendorf (UKE)
Philips Medical Systems
UMC Utrecht
Study design
Observational model
COHORT
Time perspective
RETROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
18 Years to No maximum
Healthy volunteers
No
Inclusion criteria
* Available MR angiography (MRA) scans of the Circle of Willis * Presence of at least one intracranial aneurysm (IA) * Availability of follow-up imaging or clinical records indicating stability, growth, or rupture
Exclusion criteria
* Imaging of insufficient quality for segmentation or analysis * Lack of follow-up data
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Discriminative performance of AI model for aneurysm instability prediction | At time of model validation; Up to 36 months from study start | Model performance measured by C-statistic (AUC) with a target threshold of ≥ 0.80. Instability is defined as aneurysm growth and/or rupture during follow-up. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Sensitivity and specificity of AI model | At time of model validation; Up to 36 months from study start | Targets: Sensitivity ≥ 80% and specificity ≥ 50%, based on optimal cut-off derived from ROC analysis. |
| Reduction in inter- and intra-observer variability | During vignette study in clinical setting; During final 6 months of study | Measured as reduction in variation in risk classification between and within clinicians. Target: ≥ 50% reduction. |
| Clinician adherence to AI suggestions | During vignette study; During final 6 months of study | Proportion of cases where interventional radiologists follow AI risk estimates in their decision-making. Target adherence: ≥ 80%. |
| Reduction in clinical decision-making time | During vignette study; During final 6 months of study | Time required for clinical decision-making with AI compared to without AI. Target reduction: ≥ 25%. |
| User satisfaction with AI tool | Immediately after using the AI tool in vignette study; During final 6 months of study | Self-reported satisfaction measured using a 5-point Likert scale (1 = very dissatisfied, 5 = very satisfied). Higher scores indicate greater satisfaction. Target: average score ≥ 4. |
Countries
Netherlands
Outcome results
None listed