1037 Improved Prediction of Aneurysm Growth With Deep Learning and Clinical Data Integration

INTRODUCTION: Growth of unruptured intracranial aneurysms (IAs) is a significant risk factor for rupture. Predicting IA growth remains challenging but could facilitate optimized imaging surveillance and timely intervention. METHODS: A retrospective review identified patients with unruptured IAs and over one year of imaging follow-up between 2012 and 2021. Aneurysm growth was defined as a ≥1 mm/year increase in diameter or the emergence of morphological changes such as de novo bleb formation. Cubic MRA image patches (64 × 64 × 32 voxels) were extracted, and a residual 3D convolutional neural network (3D-CNN) was trained to classify aneurysm growth. The model was evaluated using a hold-out test set and compared with the conventional ELAPSS score. We also investigated whether integrating patient background data (age, sex, comorbidities, etc.) improved predictive accuracy. RESULTS: A total of 436 patients with 579 aneurysms were included (mean age 62; 73% female). The average aneurysm diameter was 4.3 mm, and 22% of aneurysms exhibited growth during a mean follow-up of 1947 days. The ELAPSS score showed moderate predictive performance (AUC = 0.67, 95% CI: 0.61–0.72). In contrast, our AI model achieved an AUC of 0.793 (95% CI: 0.660–0.909), sensitivity of 72.7%, and specificity of 85.1%. Incorporation of patient background factors further improved the AUC to 0.838 and sensitivity to 93.3%. CONCLUSIONS: Our AI-based prediction system outperformed the ELAPSS score in forecasting IA growth. The integration of clinical background data further enhanced model accuracy, supporting its potential for clinical implementation in aneurysm monitoring.