Short summary

This study aims to establish an artificial intelligence model to aid in the description and diagnosis of unruptured intracranial aneurysms, and to estimate the rupture risk of intracranial aneurysms. The project cosists of 3 phases: 1) Algorithm construction phase. 2) Algorithm test and learning phase. 3) Algorithm validation phase. The algorithm is trained on patient journal data and CTA og MRA scans of the brain, and validated on image material from patients without intracranial aneurysm

Rationale

The prevalence of unruptured intracranial aneurysms (UIAs) in the Danish population is around 3-5%. An aneurism's yearly cumulative risk of rupture is estimated 1-2%, but the risk is highly influenced by factors such as aneurysm location and size, size and morphology changes, patient age and lifestyle, and comorbidities.

Aneurysm rupture has potentially devestating consequences. Aneurysm rupture is responsible for 85% of non-traumatic subarachnoid hemorrhages, which are associated with a 30-day mortality rate of 35%. Additionally, 65% of patients suffer permanent neurological deficits.

The current guidelines from European Stroke Organization (ESO) on management of UIAs are based on little - and low - evidence. UIA rupture risk assesment is evaluated by scoring systems such as UIATS and PHASES, which are based on correlative evidence from datamining large study populations and questions the applicability.

UIA rupture risk assessment is applied when determining potential treatment of the aneurysm, though the shortcomings and risks of the treatment procedures must also be considered.

In recent years, artificial intelligence has been investigated for its potential in clinical work. Studies have investigated the use of AI in aneurysm growth assessment, identification of aneurysm ostium and morphological descriptors, and aneurysm recurrence risk after endovascular treatment, but they have yet to describe an algorithm capable of a complete rupture risk assessmen.

An AI tool could greatly improve the precision of aneurysm description and outcome prediction. This study aims to develop an AI model, which can assist in detection, description, and rupture predic-tion of unruptured intracranial aneurysms.

Description of the cohort

This retrospective cohort includes patients from Region of Southern Denmark, who between 1994 to 2021 have been diagnosed with or followed up due to one or more intracranial aneurysms without rupture. Patients must have at least one follow-up scan of the aneurysm before eventual rupture.

For validation of the algorithm, the study includes a control group of patients without aneurysms, who have had a CTA- or MRA-scan of the brain due to any other indication.

Data and biological material

Phase 1 - Algorithm construction: patient journal data, demograpic data, diagnosis and treatment data, and radiological images from MRA or CTA scans are collected.

Phase 2 - Algorithm training: Includes the same patient category and data as in phase 1, only for a larger number of patients.

Phase 3 - Algorithm validation: Includes the same patient cagegory and data as in phase 1 and 2, as well as radiological images and image descriptions of CTA- or MRA-scans from patients in the control group. Only image material is included from patients in the control group.