Abstract TP240: Artificial Intelligence Outperforms Radiologists in Detecting Aneurysm Growth

Introduction: Cerebral aneurysm growth is a well-established indicator of instability and rupture risk. Growth is commonly assessed with linear measurements on CT or MR angiography, which may be prone to measurement error and do not account for aneurysm morphology or changes in surface area. We hypothesized that automated measurement using artificial intelligence software would more accurately and precisely detect aneurysm growth. Methods: We conducted a retrospective cohort study of patients with unruptured intracranial aneurysms who underwent serial CTA or MRA imaging from 1/27/2003 to 5/1/2024. De-identified imaging was processed by RAPID Aneurysm (iSchemaView, Menlo Park, CA), a semi-automated AI platform that performs segmentation-based 3D modeling and measures maximum linear dimension (MLD), volume and surface area. Aneurysm size changes were independently assessed in each comparative method by calculating percent change in MLD. Aneurysm growth was based on established criteria, using thresholds that account for measurement error: >10% for aneurysms <2mm, >5% for 2-10mm, and >2% for >10mm. Discrepancies between RAPID and neuroradiologists were adjudicated in an independent and blinded manner by a neurointerventionalist with 12-years of experience, which served as the gold standard. Results: 44 patients with 78 unique aneurysms and 146 total follow-up MLD measurements on serial imaging met inclusion criteria. 33 patients (75%) were female. Median age was 60 (IQR: 47.5-65). Mean and median follow-up intervals were 2.1 and 1.3 years. Aneurysms ranged from 2.93 to 18.15 mm in MLD on RAPID measurements. 28 MLD measurements exhibited growth, RAPID detected 27 of these whereas neuroradiology detected 14. 118 MLD measurements were stable or shrinking, RAPID correctly identified this in 111 measurements versus 112 for neuroradiology. RAPID demonstrated high diagnostic performance, with a sensitivity of 0.96 and specificity of 0.94, compared with neuroradiology, which had a sensitivity of 0.50 and specificity of 0.95. Conclusion: In conclusion, RAPID Aneurysm outperformed neuroradiologists by detecting true linear growth with markedly higher sensitivity while maintaining comparable specificity, demonstrating strong potential to improve longitudinal aneurysm monitoring and may better capture growth dynamics relevant to rupture risk.