Segmentation is one of the most important and difficult procedures in medical image analysis and in its clinical applications; and blood vessels are especially difficult to segment. In this paper, we propose an isosurface-based level set framework to extract vasculature tree from magnetic resonance angiography (MRA) volumes. First, we process the extracted isosurface of MRA via the surface normal vectors; then use canny edge detection to compute image-based speed function for level set evolution to refine the processed isosurface for the exact segmentation. Results on cases demonstrate it is the mostly accuracy and efficiency of the approach.