Abstract DP23: Risk factor analysis and novel prediction system for intracranial aneurysm growth based on artificial intelligence

Abstract

Objective: The growth of unruptured intracranial aneurysms (IAs) is regarded as a critical precursor to aneurysmal rupture. Accurately predicting aneurysm growth is crucial for appropriate therapeutic interventions to prevent rupture in high-risk aneurysms. The UCAS Japan score has been widely used for rupture risk assessment in Japan; however, its relationship to aneurysm growth is unclear. The present study aimed to examine whether the UCAS Japan score can accurately predict aneurysm growth and to develop a novel prediction system using artificial intelligence (AI)-based machine learning. Methods: We retrospectively analyzed 2,399 unruptured IAs from our single institutional database between 2012 and 2021. Cases with low-quality MRI images or short follow-up periods within 12 months were excluded, resulting in 725 included IAs. IAs with 1mm or more growth during follow-up were categorized as the growth group and compared with the non-growth group. Univariate and multivariate analyses were performed based on UCAS Japan scores and possible risk factors regarding patient characteristics and aneurysmal morphology. AI-based prediction model using XGBoost method was developed and compared with the logistic analysis model and conventional ELAPSS scoring system. Results: A total of 150 aneurysms were classified into the growth group and 525 into the non-growth group. The average age was 63.9±11.7 years, with 74.9% female. Univariate analysis showed significant differences between the two groups in age (65.1 vs 63.3, p=0.01), hypertension (63.3% vs 52.3%, p=0.02), maximum diameter (6.5 mm vs 4.3 mm, p<0.01), and daughter sac presence (53.3% vs 10.4%, p<0.01). The UCAS Japan score was only slightly higher in the growth group (4.90 vs 4.70, p<0.01). Multivariate analysis identified daughter sac presence (OR 8.27, 95%CI 5.16–13.30) and family history of SAH (OR 2.52, 95%CI 1.37–4.64) as independent risk factors. The AI-based model showed strong predictive performance (AUC 0.91, accuracy 91%, sensitivity 71%, specificity 96%), surpassing the ELAPSS score (AUC 0.67) and logistic regression (AUC 0.84, p<0.01). Conclusion: Multivariate analysis identified family history of SAH, daughter sac presence, and aneurysm size as independent risk factors for IA growth. We developed an AI-based powerful prediction system, which potentially change IA management dramatically.

Ähnliche Arbeiten