anatomical mri module
DESCRIPTION
MNTP Trainee: Georgina Vinyes Junque , Chi Hun Kim Prof. James T. Becker Cyrus Raji , Leonid Teverovskiy , and Robert Tamburo. Anatomical MRI module. Voxel -Based Morphometry (VBM). Structural differences based on Voxel -wise comparision Advantages - PowerPoint PPT PresentationTRANSCRIPT
Anatomical MRI mod-ule
MNTP Trainee: Georgina Vinyes Junque, Chi Hun KimProf. James T. BeckerCyrus Raji, Leonid Teverovskiy, and Robert Tamburo
Voxel-Based Morphometry (VBM)
Structural differences based on Voxel-wise com-parision
Advantages Automated, Un-biased, Whole brain analysis compared
to Manual ROI tracing Well established and Widely used over the past decade Results are biologically plausible and replicable We know the LIMITATIONS
Overview
Voxel-Based Mor-phometry Bias Field Correction Skull Stripping Spatial Normalization to
Template Tissue Segmentation Modulation SmoothingVoxel-wise statistical tests
Preprocessing
Methods
MRI sequence T1 (MPRAGE) 3T Siemens TrioTim Slices: 160; thickness 1.2mm Voxel size: 1 x 1 x 1.2 mm TE: 2.98; TR: 2300
Software SPM2 & SPM5 (Wellcome Trust
Centre for Neuroimaging) VBM2 toolbox (Gaser et al,
http://dbm.neuro.uni-jena.de/) N3 algorithm Brain Extraction Tool in FSL Watershed algorithm in
FreeSurfer
Subjects Multicenter AIDS Cohort
Study (MACS) 53 males Age: 50.2 +- 4.4
Statistical Analysis Gray matter Volume dif-
ferences in Drug users vs. Non-
Drug users
MRI Bias Field Correc-tion
Origi-nalImage
Cor-rectedImage Corrected Bias field
= Original – Corrected image
Software: N3 (Nonparametric Nonuniform intensity Normalization)
N3
Experiment 1. Adding ’Known’ Bias Field
Known Bias Field
+
Successful Removal ofKnown Bias field
N3
Experiment 2. ’Repetition’ of Bias Field Correction
Original im-age
Corrected imageAfter 5th repetition
< Amount of Corrected Bias Field over N3 Repetition >
1 2 3 4 50
1
2
3
4
5
6
7
Mean
# of repetition
Mea
n S
igna
l Int
ensi
ty o
f C
orre
cted
Bia
s F
ield
N3
N3
N3
N3
N3
Skull Stripping
Software Brain Extraction Tool (BET; v2.1 in FSL software package) Watershed algorithm in FreeSurfer software package v5.1.0
BET default setting (1 min) Watershed default setting (30 min)
Optimization of Parameters (2min)
Skull Stripping: Brain Extraction via Deformable Registration
Teverovskiy, 2011, OHBM, Poster Presentation
Spatial Normalization to Template
1. Customized template Recommended in special populations
(Eg: babies or the elderly).
2. Standardized template Better comparison with similar studies
using the same template. Eg. MNI: 152 brains, mean age 25, female 43%
http://dbm.neuro.uni-jena.de/vbm/vbm2-for-spm2/creating-custom-ized-template/
Fitting each individual brain into the same brain template, To compare regional differ-ences between groups
Effect of 3 Different Tem-plate on Statistical ResultsMACS template Default-MNI templateCustomized template
Glass brains, showing reduced grey matter volume in drug users compared to non-drug users, at 0.01 Uncorrected level
Segmentation into 3 Tissue Types
2. Tissue Probability Map
http://dbm.neuro.uni-jena.de/vbm/segmen-tation/
1. Signal Intensity of Voxel
Grey Mater Segmen-tation
CSFSegmenta-
tion
White Mater
Segmenta-tion
Modulation
It’s recommended if you are more interested in volume changes than
differences in concentration (or density)
http://dbm.neuro.uni-jena.de/vbm/segmentation/modu-lation/
Recovering volume information which was lost by spatial normalization process.
It can be thought as atrophy correction.
Effects of Modulation on Results
Modulated:Changes in GM volume
Unmodulated:Changes in GM density
Glass brains showing reduced grey matter in drug users compared to non-drug users, at 0.01 Uncorrected level
Smoothing
Intensity of every voxel is replaced by the weighted average of the surrounding vox-els. Larger kernel size, more surrounding voxels
Make distribution closely to Gaussian field model
Increase the sensitivity of tests by reduc-ing the variance across subjects
Reduce the effect of misregistration
Effect of Different Smoothing Kernels
Glass brains showing reduced grey matter volume in drug users compared to non-drug users, at 0.01 Uncorrected level
5 mm 10 mm 15 mm
Conclusion
There’s a lot of options in processing that can affect data and results.
We have to undertand what we are doing in every step to better adjust options to our sample study.
Since these techniques have several pitfalls, we have to carefully inter-pret published results.
Thank You
Prof. James T. BeckerTA: Cyrus Raji, Leonid Teverovskiy,
Robert TamburoProf. Seong-Gi Kim & Prof. Bill EddyTomika Cohen, Rebecca ClarkFellow MNTPers!