Accurate Brain Vessel Segmentation in T1-Weighted Mri Based on Unetr: Improving Neurosurgical Planning Publisher



Gholizadeh F ; Rahmani M ; Pourrashidi A ; Najafzadeh E ; Farnia P ; Ahmadian A
Source: 32nd National and 10th International Iranian Conference on Biomedical Engineering, ICBME 2025 Published:2025

Abstract

Preoperative planning for brain tumor surgeries is highly challenging and requires precise identification of vascular anatomy to minimize the risk of complications. While T1-weighted contrast-enhanced (T1CE) MRI is routinely used for preoperative assessment, automated vessel segmentation from these scans remains a significant challenge. The absence of reliable vessel maps can disrupt surgical workflows and may compromise patient safety, especially in settings where specialized angiographic imaging is not available. In this study, we propose a transformer-based UNETR model that leverages global contextual information to address the complexity of brain vessel segmentation. After standardized preprocessing, the model was trained and validated on 30 expert-annotated T1CE MRI scans. The approach achieved high performance, with a Dice score of 87%, IoU of 0.98, sensitivity of 0.99, and specificity of 0.99, showing strong capability in detecting both major vessels and smaller vascular branches. These findings highlight the potential of attention-based architectures to enhance routine clinical imaging by providing accurate vessel maps directly from standard MRI sequences already acquired for tumor evaluation. Such a framework could support safer and more efficient preoperative planning without requiring additional imaging resources. © 2025 IEEE.