Medical-AI trial: detecting brain aneurysms from head MRI (TOF-MRA) with AI (RDX-Aneurysm) — testing performance against misses (NCT07655635)

A brain (intracranial) aneurysm is a balloon-like bulge in the wall of a brain vessel; most are asymptomatic, but a rupture causes the serious event of subarachnoid hemorrhage. Found before rupture, it can be watched or treated preventively. TOF-MRA, a form of brain MRI that depicts vessels from blood flow without contrast, is used to find aneurysms, but small or ambiguous ones are easily missed.

That is where computer-assisted detection (CADe) software, which analyzes the image and points to suspected aneurysm sites, comes in.

This study evaluates the performance of that CADe software, RDX-Aneurysm, as a retrospective, multicenter, case-control study.

Per the registry summary, it evaluates the standalone performance of RDX-Aneurysm in detecting and marking suspected saccular (sac-like) intracranial aneurysms on adult head TOF-MRA. The evaluation includes about 550 exams: roughly 250 positive exams with at least one reference-confirmed saccular aneurysm 3 mm or larger, and roughly 300 negative exams with no confirmed aneurysm. The primary outcome is lesion-level sensitivity for saccular aneurysms 3 mm or greater.

Missing a brain aneurysm can lead to a preventable rupture and is an important challenge in radiology reading. If AI can mark suspicious sites, it could act as a second set of eyes to reduce reader misses. Splitting positive and negative exams in a case-control design and setting lesion-level sensitivity (how well it avoids missing lesions) as the primary outcome aptly measures a detection software.

That said, this study evaluates standalone performance retrospectively; it does not establish replacement of the reading physician, clinical adoption, or prevention of rupture.