Anatomical segmentation of the human brain: comparative assessment of two automatic methods

Abstract

Magnetic Resonance Imaging (MRI) is a robust and versatile imaging modality and an integral component of a lot of studies, especially when performing quantitative analysis. MRI is the preferred method of imaging the brain because of its excellent soft tissue contrast. Accurate segmentation of the brain into its anatomical regions enables accurate quantitative analysis of the brain. Three software programs that perform automatic anatomical segmentation of the human brain are FreeSurfer, FastSurfer and MAPER. The purpose of this study was to use FreeSurfer as a baseline, and to investigate how well FastSurfer and MAPER segmentations conform to FreeSurfer’s outputs on the same dataset. 185 T1-weighted 3D MR images from the IXI Dataset were segmented using FreeSurfer, FastSurfer, and MAPER. Default training checkpoints were used for FastSurfer and FreeSurfer outputs of the IXI Dataset, along with corresponding brain MR images, were used as a source atlas for MAPER. The FastSurfer and MAPER segmentations were then compared with the FreeSurfer segmentations using the Jaccard Similarity Coefficient. MAPER performed better than FastSurfer at replicating FreeSurfer-conform outputs for subcortical regions. MAPER and FastSurfer performed similarly for the cortical regions.