Voxelbased Morphometric Analysis Martinos Center for Biomedical Imaging
Voxel-based Morphometric Analysis Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA, United States 1
Outline • Voxel-based Morphometry • Surface-based Analysis • Comparisons 2
Voxel-based Morphometry 3
Anatomical Changes GM Subject 1 88 yo Subject 2 19 yo 4
Voxel-based Morphometry (VBM) How do the sizes of gray/white matter and CSF structures change between subjects/populations? GM WM CSF Subject 1 88 yo Subject 2 19 yo 5
Voxel-based Morphometry (VBM) How to define a volume without defining a boundary? How to compare regions without defining a region? WM CSF 6
Non-linear Spatial Normalization Subject 1 Subject 2 (Target) Subject 1 Subject 2 CSF • Eyes move closer together • Lips curl and get wider 7
Non-linear Spatial Normalization Subject 1 Subject 2 (Target) Subject 1 Subject 2 CSF • Eyes move closer together • Lips curl and get wider 8
Keep Track of Changes in Size CSF • Voxel on the left side of mouth does not change size • Voxel on the right gets much larger • Eyes do not change size 9 • Quantification: Gray Matter “Density”
Jacobian Subject 1 Subject 2 (Target) Jacobian Subject 2 (Target) CSF • Map of change in volume at each voxel in target space • Eyes are 0 • Left side of lips are 0 (black) • Right side of lips are yellow (expansion) 10
Normalization and Segmentation Individual T 1 (Template Space) Individual T 1 Group Template (Target) Jacobian Expansion Compression Spatial Normalization Segmentations CSF “Unified Segmentation”, Ashburner and Friston, NI, 2005 “Optimized VBM” Good, et al, NI, 2001. Douaud, et al, Brain. 2007. WM GM Values between 0 and 1 “Density”, Partial Volume Note: in FSL, Segmentations computed in 11 native space
Modulation and Smoothing (Template Space) GM Segmentation (Concentration) Multiply 3 D Smooth GM Density Jacobian 12
Aging Gray Matter Volume Study N=40 Statistical Maps (SPM 8/VBM 8) p<. 01 GM Density Positive Age Correlation Negative Age Correlation 13
Surface-based Analysis 14
Surface-based Analysis: Cortex • Outer layer of gray matter • White/Gray Surface • Pial Surface 15
Cortical Thickness pial surface • Distance between white and pial surfaces along normal vector. • 1 -5 mm 16
Individual. Thickness Maps 18 yo 46 yo 88 yo Male Female Salat, et al, 2004, Cerebral Cortex 17
A Surface-Based Coordinate System Common space for group analysis (like Talairach). Fischl, et al, 1998, NI. 18
Surface Spatial Smoothing • 5 mm apart in 3 D • 25 mm apart on surface! • Kernel much larger • Averaging with other tissue types (WM, CSF) • Averaging with other functional areas 19
Aging Thickness Study N=40 Positive Age Correlation p<. 01 Negative Age Correlation 20
VBM and Thickness vs Age VBM (SPM 8/VBM 8) Thickness (Free. Surfer 5. 0) p<. 01 21
Comparisons • Thickness does not require modulation • False positive rates are much higher in VBM because of Jacobian modulation (same for surfacebased when area or volume are used; Greve and Fischl, 2017) • Thickness is independent of registration. • VBM – harder to interpret because “density” is a mixture of thickness, surface area, gyrification, registration, volume-based smoothing, and intensity. • VBM allows subcortical analysis Young (20) Old (80) 22
Which is better? • Still an open question • Voets, et al, 2008, NI – mixed results • Hutton, et al, 2009, NI – voxel-based cortical thickness (VBCT) was more sensitive to aging than VBM • False Positive Rate considerations 23
VBM Software • Statistical Parametric Mapping (SPM) • www. fil. ion. ucl. ac. uk/spm • uses the VBM toolbox • dbm. neuro. uni-jena. de/vbm • FMRIB Software Library (FSL) • www. fmrib. ox. ac. uk/fsl 24
Thanks! 25
26
VBM Summary • Spatial Normalization • Volume = Jacobian X Segmentation • Strengths • Cortical and subcortical • Gray Matter, White Matter, CSF • Easy to use • Weaknesses • Volume metric derived from normalization • Sensitive to registration and segmentation errors • Segmentation is atlas-dependent 27
Surface-based Summary • Thickness (can also use area and volume) • Strengths • Surface-based Normalization • Surface-based Smoothing • Thickness independent of normalization • Surface extraction atlas-independent • Weaknesses • No subcortical, White Matter, or CSF • More complicated to analyze 28
- Slides: 28