The effects of white matter disease on the accuracy of automated segmentation

Abstract

Automated segmentation of the brain is challenging in the presence of brain pathologies such as white matter hyperintensities (WMH). A late-life depression population was used to demonstrate the effect of WMH on brain segmentation and normalization. We used an automated algorithm to detect WMH, and either filled them with normal-appearing white-matter (NAWM) intensities or performed a multi-spectral segmentation, and finally compared the standard approach to the WMH filling or multi-spectral segmentation approach using intra-class correlation coefficients (ICC). The presence of WMH affected segmentations for both approaches suggesting that studies investigating structural differences in populations with high WMH should account for WMH. We also investigated how functional data contrasts are affected using normalization between the standard compared to fill and multi-spectral approach. We found that the functional data was not affected. While replication with a larger sample is needed, this study shows that WMH can significantly affect the results of segmentation and these areas are not limited to those affected by WMH. It is clear that to study gray matter differences that some correction should be made to account for WMH. Future studies should investigate which methods for accounting for WMH are most effective.

Amyloid-Beta Deposition is Associated with Increased Medial Temporal Lobe Activation during Memory Encoding in the Cognitively Normal Elderly

“…We have used SPM’s methods to perform this analysis. Our group has shown that the effect of WMH is negligible on the parameter estimates in fMRI analyses (40). However, due to the limited resolution of the fMRI data, the specificity of these procedures becomes less critical.…”

Voxel-wise hemispheric Amyloid Asymmetry and its association with cerebral metabolism and grey matter density in cognitively normal older adults

“…Sagittal whole brain 3D magnetization prepared rapid-acquisition gradient MRI then underwent standard preprocessing using Statistical Parametric Mapping (SPM12) 21 toolbox in MATLAB2016b (Mathworks) and structural sequences were coregistered to MPRAGE, bias-corrected, and segmented into 6 classes 22 . Due to presence of white matter hyperintensities, we adjusted the number of Gaussians used to identify white matter to 2 to improve identification of gray and white matter 23 . We used Diffeomorphic Anatomical Registration using Exponentiated Lie Algebra (DARTEL) to generate a study template 24 .…”

Regional SUV quantification in hybrid PET/MR, a comparison of two atlas-based automatic brain segmentation methods

“…However, the evaluation of effects of different automatic segmentation methods on PET SUV quantification is lacking. To label brain regions for group analysis, automatic segmentation RSIAT always fits the subjects' images to a common reference space [24,25], and PET quantification is performed in the transformed space, which fits the individual image data to adapt to the common reference frame and probably influences the accuracy of regional SUVs [26].…”