Brain Volume Segmentation Outliers Correction in Structural MRI Images

The skull stripping procedure is an important image preprocessing step commonly applied in many neuroscience studies. Even though several efforts have been made in order to create robust brain extraction algorithms, minor segmentation errors still remain, often requiring manual refinement. In this study, an automatic Brain Volume Refinement (BVeR) method is proposed. The method interprets segmentation outliers as local interference in brain tissue signal contrast, offering a suitable solution for external brain boundary adjustment of structural T1 and T2 weighted MRI. Two publicly available structural MRI image datasets of healthy adults and two commonly used brain extraction methods (BET and FreeSurfer) were used for evaluation. Quantitative segmentation evaluation for accuracy and reproducibility were applied to evaluate the performance of BVeR, showing that the average brain volume refinement showed a significant improvement (p < 0.001) in most metrics. In conclusion, the BVeR method offers an automatic alternative to the manual correction often requested in brain MRI studies, in which it considerably reduces human errors and processing time.