MULTI-PARAMETRIC 3D MAGNETIC RESONANCE IMAGE BRAIN TUMOR SEGMENTATION METHOD

摘要

An automatic brain tumor segmentation method for a multi-parametric 3D magnetic resonance image, including: classifying each voxel in an image and acquiring the probability that the image belongs to a brain tumor area; extracting multi-scale structure information about the image; constructing a multi-scale brain tumor probability image based on an initial brain tumor probability image and the multi-scale structure information about the image; determining a significant tumor area based on the multi-scale brain tumor probability image; acquiring a robust initial tumor segmentation tag based on tumor probability information about an initial image and the significant tumor area; and using a tag propagation method based on the graph theory to separate the brain tumor area. The present invention acquires robust initial tumor segmentation tag information which is reliable in statistics, compact in space and has enough voxels, which is advantageous for the accuracy and reliability of the tumor segmentation result. The segmentation method based on tag propagation reduces the influence on the segmentation result by inter-individual image grayscale difference and insufficient training statistics information to a certain extent, improving the stability of tumor segmentation.