A preprocessing method for magnetic resonance images of head to improve the performance of brain extraction tools

Abstract

An automated preprocessing method to improve the performance of brain segmentation with Brain Extraction Tools (BET) is proposed in this paper. It is an automatic decision method of optimal cutting plane of non-brain tissues. This method can be easily integrated into BET and can significantly improve the brain segmentation result for troublesome images. Some clinical 3D T1-weighted MRI data of glioma patients have been used to evaluate this method. Both BET with and without this preprocessing method are carried out, and the results are compared with the “gold-standard” segmented by experts. Two criteria are used to judge the performance, overlay rate and extra rate, where extra rate indicates the ratio of the amount of residual non-brain tissues after segmentation to total non-brain tissues. In the case where no neck area exists the patient head data, both of the BET with and without this preprocessing give the same result. But in the case where the neck is partly included, this proposed preprocessing achieves an average overlay rate of 97.28%, greater than 97% provided by non-preprocessing; and in term of extra rate, the BET with this preprocessing can reach 5.08% that much lower than 29.3% of that without preprocessing.