Research involving animals contributes to our comprehension of the pathology and progression of cerebrovascular diseases. Utilizing ultra-high field 3D time-of-flight magnetic resonance angiography (TOF-MRA) has facilitated noninvasive, high-resolution imaging of the vasculature in mice. Despite this advancement, there is currently a lack of tools for segmenting vasculature in 3D TOF-MRA images of mice.In this study, we introduce a novel approach employing an attention network for automatic segmentation of cerebral vessels in 11.7T TOF-MRA images of mice. The proposed method was trained and evaluated using 34 TOF-MRA volumes. In contrast to other state-of-the-art segmentation networks, our method demonstrated superior completeness in capturing cerebrovascular structures. Compared with manual labeling, the proposed method achieved a Dice similarity coefficient of 85.50%. This methodology can serve as an effective tool for angiography analysis in pre-clinical studies of cerebrovascular diseases.