A Bayesian model of shape and appearance for subcortical brain

FMRIB Centre, Department of Clinical Neurology, University of Oxford, Oxford, UK.
NeuroImage (Impact Factor: 6.36). 02/2011; 56(3):907-22. DOI: 10.1016/j.neuroimage.2011.02.046
Source: PubMed

ABSTRACT Automatic segmentation of subcortical structures in human brain MR images is an important but difficult task due to poor and variable intensity contrast. Clear, well-defined intensity features are absent in many places along typical structure boundaries and so extra information is required to achieve successful segmentation. A method is proposed here that uses manually labelled image data to provide anatomical training information. It utilises the principles of the Active Shape and Appearance Models but places them within a Bayesian framework, allowing probabilistic relationships between shape and intensity to be fully exploited. The model is trained for 15 different subcortical structures using 336 manually-labelled T1-weighted MR images. Using the Bayesian approach, conditional probabilities can be calculated easily and efficiently, avoiding technical problems of ill-conditioned covariance matrices, even with weak priors, and eliminating the need for fitting extra empirical scaling parameters, as is required in standard Active Appearance Models. Furthermore, differences in boundary vertex locations provide a direct, purely local measure of geometric change in structure between groups that, unlike voxel-based morphometry, is not dependent on tissue classification methods or arbitrary smoothing. In this paper the fully-automated segmentation method is presented and assessed both quantitatively, using Leave-One-Out testing on the 336 training images, and qualitatively, using an independent clinical dataset involving Alzheimer's disease. Median Dice overlaps between 0.7 and 0.9 are obtained with this method, which is comparable or better than other automated methods. An implementation of this method, called FIRST, is currently distributed with the freely-available FSL package.

Download full-text


Available from: Brian Patenaude, Apr 24, 2014
  • Source
    • "This process was repeated for the PRc and PHc masks. FSL FIRST module (Patenaude et al. 2011) was used to automatically segment the HC. Using histologically derived guidelines, the automatically segmented HC mask was manually cleaned by all three raters. "
    [Show abstract] [Hide abstract]
    ABSTRACT: This study examined the reliability of high angular resolution diffusion tensor imaging (HARDI) data collected on a single individual across several sessions using the same scanner. HARDI data was acquired for one healthy adult male at the same time of day on ten separate days across a one-month period. Environmental factors (e.g. temperature) were controlled across scanning sessions. Tract Based Spatial Statistics (TBSS) was used to assess session-to-session variability in measures of diffusion, fractional anisotropy (FA) and mean diffusivity (MD). To address reliability within specific structures of the medial temporal lobe (MTL; the focus of an ongoing investigation), probabilistic tractography segmented the Entorhinal cortex (ERc) based on connections with Hippocampus (HC), Perirhinal (PRc) and Parahippocampal (PHc) cortices. Streamline tractography generated edge weight (EW) metrics for the aforementioned ERc connections and, as comparison regions, connections between left and right rostral and caudal anterior cingulate cortex (ACC). Coefficients of variation (CoV) were derived for the surface area and volumes of these ERc connectivity-defined regions (CDR) and for EW across all ten scans, expecting that scan-to-scan reliability would yield low CoVs. TBSS revealed no significant variation in FA or MD across scanning sessions. Probabilistic tractography successfully reproduced histologically-verified adjacent medial temporal lobe circuits. Tractography-derived metrics displayed larger ranges of scanner-to-scanner variability. Connections involving HC displayed greater variability than metrics of connection between other investigated regions. By confirming the test retest reliability of HARDI data acquisition, support for the validity of significant results derived from diffusion data can be obtained.
    Brain Imaging and Behavior 07/2015; DOI:10.1007/s11682-015-9425-1 · 4.60 Impact Factor
  • Source
    • "We conducted an additional analysis to identify whether there were areas of the hippocampus whose local volume differed between groups. The method was based on FSL FIRST's vertex analysis (Patenaude et al., 2011), but modified to allow incorporation of manual edits on the hippocampal structure (following suggestions by Jenkinson, 2014). This analysis was not based on comparison of mesh-based segmentations of the hippocampus but rather on comparisons of the outer envelope of participants' hippocampi in common space. "
    [Show abstract] [Hide abstract]
    ABSTRACT: We studied a group of verbal memory specialists to determine whether intensive oral text memory is associated with structural features of hippocampal and lateral-temporal regions implicated in language processing. Professional Vedic Sanskrit Pandits in India train from childhood for around 10 years in an ancient, formalized tradition of oral Sanskrit text memorization and recitation, mastering the exact pronunciation and invariant content of multiple 40,000-100,000 word oral texts. We conducted structural analysis of grey matter density, cortical thickness, local gyrification, and white matter structure, relative to matched controls. We found massive gray matter density and cortical thickness increases in Pandit brains in language, memory and visual systems, including i) bilateral lateral temporal cortices and ii) the anterior cingulate cortex and the hippocampus, regions associated with long and short-term memory. Differences in hippocampal morphometry matched those previously documented for expert spatial navigators and individuals with good verbal working memory. The findings provide unique insight into the brain organization implementing formalized oral knowledge systems. Copyright © 2015. Published by Elsevier Inc.
    NeuroImage 07/2015; DOI:10.1016/j.neuroimage.2015.07.027 · 6.36 Impact Factor
  • Source
    • "First, non-brain tissue was removed from the images with FSL BET (Brain Extraction Tool, Smith, 2002), using the default parameters and no options. Next, for cortical and subcortical tissue classification, we used FSL FAST (FMRIB's Automated Segmentation Tool, Zhang et al., 2001) and FSL FIRST (FMRIB's Linear Image Registration Tool, Patenaude et al., 2011) with the default parameters and no options. The experiment was repeated twice in each execution condition to ensure that no inter-run differences were present. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Neuroimaging pipelines are known to generate different results depending on the computing platform where they are compiled and executed. We quantify these differences for brain tissue classification, fMRI analysis, and cortical thickness (CT) extraction, using three of the main neuroimaging packages (FSL, Freesurfer and CIVET) and different versions of GNU/Linux. We also identify some causes of these differences using library and system call interception. We find that these packages use mathematical functions based on single-precision floating-point arithmetic whose implementations in operating systems continue to evolve. While these differences have little or no impact on simple analysis pipelines such as brain extraction and cortical tissue classification, their accumulation creates important differences in longer pipelines such as subcortical tissue classification, fMRI analysis, and cortical thickness extraction. With FSL, most Dice coefficients between subcortical classifications obtained on different operating systems remain above 0.9, but values as low as 0.59 are observed. Independent component analyses (ICA) of fMRI data differ between operating systems in one third of the tested subjects, due to differences in motion correction. With Freesurfer and CIVET, in some brain regions we find an effect of build or operating system on cortical thickness. A first step to correct these reproducibility issues would be to use more precise representations of floating-point numbers in the critical sections of the pipelines. The numerical stability of pipelines should also be reviewed.
    Frontiers in Neuroinformatics 04/2015; 9(12). DOI:10.3389/fninf.2015.00012
Show more