%0 Journal Article
%T Validation of Automated White Matter Hyperintensity Segmentation
%A Sean D. Smart
%A Michael J. Firbank
%A John T. O'Brien
%J Journal of Aging Research
%D 2011
%I Hindawi Publishing Corporation
%R 10.4061/2011/391783
%X Introduction. White matter hyperintensities (WMHs) are a common finding on MRI scans of older people and are associated with vascular disease. We compared 3 methods for automatically segmenting WMHs from MRI scans. Method. An operator manually segmented WMHs on MRI images from a 3T scanner. The scans were also segmented in a fully automated fashion by three different programmes. The voxel overlap between manual and automated segmentation was compared. Results. Between observer overlap ratio was 63%. Using our previously described in-house software, we had overlap of 62.2%. We investigated the use of a modified version of SPM segmentation; however, this was not successful, with only 14% overlap. Discussion. Using our previously reported software, we demonstrated good segmentation of WMHs in a fully automated fashion. 1. Introduction Magnetic resonance imaging (MRI) is now widely used in the diagnosis of diseases by doctors and is particularly useful for scanning images of the brain and detecting cerebrovascular disorders. White matter hyperintensities (WMHs) are a common finding in elderly people which are associated with vascular risk factors and an increased risk of decline in cognitive and motor function [1]. A number of methods have been used to quantify the hyperintensities to correlate to clinical data such as visual ratings, volumetric measuring, and WMHs pattern [2¨C6]. An investigation with the LADIS study cohort found that volumetric measurement was more sensitive than visual rating to detect differences in WMHs between groups with versus without memory symptoms although both volumetric measurement and visual rating detected differences in WMHs relating to age and gait disturbance [7]. Currently, there is no accepted gold standard for a fully automated WMHs segmentation program. The SPM package (http://www.fil.ion.ucl.ac.uk/spm/software/) has a widely used segmentation tool which classes brain tissue into grey, white matter, and CSF, using a combination of image intensity and a priori knowledge regarding distribution of tissue types. The default does not include information about WMHs, and these can be misclassified as grey matter [8]. Adding information regarding the a priori distribution of WMHs may help to improve the segmentation of WMHs in SPM. The study aims to investigate the ability of SPM to segment WMHs from (a) T1 weighted and (b) T1 + FLAIR images using a priori information about WMHs distribution. Results will be compared to manual segmentation of WMHs from FLAIR images, an in-house WMHs segmentation program [9], and a different
%U http://www.hindawi.com/journals/jar/2011/391783/