P03.01.b Automated Segmentation of Cerebral Arteries on Contrast-Enhanced Mri Presenting Intracranial Tumors: An Alternative for Angiograph to Facilitate Clinical Diagnostic Assessment

Abstract BACKGROUND Accurate assessment of the spatial relationship between intracranial tumors and cerebral arteries is critical for both therapeutic decision making and preoperative planning. But angiography is not among the standard examination in tumor diagnosis or follow-up. Here, since contrast-enhanced T1-weighted MRI is available for every patient, we aim to develop an automated segmentation model for cerebral arteries with it, and to investigate if the work can be used as an alternative to angiography. MATERIAL AND METHODS A deep learning model, based on the nnU-Net framework, was trained and validated using a large multi-center dataset of 1174 cases. Pixel-level ground truth annotations was performed on six major cerebral arteries. Model performance was evaluated with Dice similarity coefficient (DSC), recall, and Hausdorff distance (HD). Comparative experiments with prospective clinical validation using TOF-MRA as the gold standard were conducted. RESULTS The results suggested that the proposed model demonstrated robust performance, achieving a DSC of 0.886 ± 0.015 on the internal test set, and 0.884 ± 0.016, 0.859 ± 0.019, 0.882 ± 0.032, and 0.864 ± 0.025 on the external test set. Clinical validation confirmed that the model effectively reconstructed large arteries, covering approximately 80% of vessels identifiable on TOF-MRA, but with lower accuracy in diameter prediction limited by the quality of input. CONCLUSION It’s concluded that the automated model for cerebral artery segmentation on T1C offers a clinically feasible alternative to angiography, facilitating tumor diagnostic workflows, and reducing patient burden.