Structural MRI Flashcards
What is a voxel?
A ‘voxel’ represents a single cell in a 3D matrix. It contains a number that represents the measured signal intensity in the x-y-z-location
What is an isotropic voxel?
Voxel that has the same ‘length’ in all dimensions (e.g., 2 x 2 x 2 mm)
What is the modern spatial resolution in an MRI image?
1 mm^3
(3) Ways to ‘slice’ through the matrix in different planes:
○ Axial/ horizontal
○ Coronal
○ Sagittal
https://faculty.washington.edu/chudler/slice.html
One of the most common motion artifacts and effect of GM?
Head motion, higher head motion leads to lower grey matter volume estimates
How to mitigate motion artifacts prospectively and post hoc?
Post hoc: exclude scans with high motion; new correction approaches for DTI data
Prospective strategies: detect and account for subject motion during the acquisition itself
Clinical MRI types:
1. T1-weighted imaging
2. T2-weighted imaging
3. FLAIR
4. T1+KM
5. T2*-weighted imaging
6. Diffusion-weighted imaging (DWI)
Clinical MRI types:
1. myelin, highly fatty substances
2. liquids, namely water
3. detecting edema, multiple sclerosis
4. checking the vessels, helps detect and classify lesions
5. detecting BOLD
6. measuring movement along axon fibers of water molecules
Voxel-wise lesion-symptom mapping approaches
- Classical approach
- Lesion-defined approach
- Behaviour-defined approach - Mass-univariate statistical analysis
‘Lesion-defined’ approach:
- behavioral performance of a group of patients with a common area of injury (e.g., DLPFC) compared to that of a control group or another patient group
- good for assessing the functional roles of particular regions of interest
- BUT: loss of information if ROI contains multiple subregions that each contribute to behavior
regions outside the ROI that are part of a distributed functional network may be overlooked
‘Behavior-defined’ approach:
- patients are grouped according to whether or not they show a specific behavioral deficit
- lesion reconstructions are overlaid to find common areas and compared to lesion overlays from patients without the deficit
- effective in identifying brain regions that may contribute to a cognitive skill
- BUT: when behavioral data are continuous → cut-off needed → information reflecting varying degrees of performance can be lost
Mass univariate statistical analysis:
○ Comparing one aspect (e.g., intensity)
○ Does not require patients to be grouped by either lesion site or a behavioral cutoff a priori
○ Makes use of continuous behavioral and lesion information
○ Voxel-by-voxel analysis
- Parametric mapping
- For each voxel, patients are categorized according to whether they did or did not have a lesion affecting that voxel
- Behavioral scores are then compared to these two groups, yielding a t-statistic for each voxel
- Allows researchers to identify specific brain regions that are associated with significant differences in behavior between patients with and without lesions in those regions
Volumetry - Manual - Pros & Cons
§ Advantages:
□ Remains the gold standard
□ High biological and anatomical validity
□ Ideal for delineating structures with intricate anatomy/multiple subregions
□ Well-suited for smaller studies with focused hypotheses
§ Disadvantages:
□ Labor-intense: impossible for large studies (> 1.000 scans)
□ Requires expert anatomical knowledge
□ Requires at least two blinded tracers to avoid bias
□ Intra-rater variability
□ Inter-rater variability
Inter-protocol variability
ITK-Snap:
- Hand-selected regions.
- Interactive tool for labeling structures in 3D medical image volumes
- Free, open-source & multi-platform C++ software with binaries, widely available for OSs
Volumetry - automated segmentation - Pros and cons
§ Advantages:
□ Replaces manual segmentation for most applications
□ Substantially faster (for large datasets)
□ Higher reliability (reduce human error and subjectivity)
□ Standardization
§ Disadvantages:
□ FreeSurfer can overestimate total hippocampal volumes
□ Problems with accurately detecting boundaries between hippocampus and neighboring structures
□ Differences in segmentation outcomes with regard to age effects and hemispheric asymmetry
□ However, agreement between manual and automated approaches is continuously improving (e.g., FreeSurfer versions 6 and 7)
FSL FIRST, FreeSurfer and FSL SIENA(X) segmentation
- FSL FIRST - subcortical
- FreeSurfer - subcortical + global
- FSL SIENA(X) - global tissue volumes (whole brain), cross-sectional, FSL SIENA - longitudinal, good for estimation of atrophy
What is Voxel-based morphometry (VBM)?
- Measures local concentration (density) of grey matter
- Characterizes local differences in grey matter topography, while discounting for large-scale differences in anatomy (relative concentration of grey matter structures in the spatially normalized images)
- Align images globally and compare GM likelihood at each voxel
Steps of VBM:
- Brian extraction
- Segmentation into tissue classes (DW, WM, CSF)
- Normalization
- Smoothing
Steps of VBM: Segmentation
- Often combined with intensity normalization
- Classification commonly combination of intensity-based clustering informed by spatial priors
- Cleaning step (remove non-brain classes)
Steps of VBM: Normalization
- Linear vs. non-linear
- Linear: manipulation (e.g., rotation) is applied equally to all parts of the image
- Non-linear: manipulation is applied locally (i.e., different changes to different parts)
- Choice of template can impact the results (MNI305, MNI152, …)
Steps of VBM: Smoothing
- Smoothing makes each voxel more similar to its neighbors (weighted means)
- Typical smoothing kernels are 4-16 mm full width at half maximum (FWHM)
- Why? Parametric tests assume Gaussianity of residuals, smoothing helps satisfy this assumption
- Spatial normalization is not perfect, and smoothing helps accommodate inter-individual differences in local anatomy
- Smoothing makes the analysis sensitive to the kernel size, such that very small differences are disregarded
What is partial volume effect?
Caveat in VBMl leading to interpretational uncertainty. Nearby tissues (like WM, blood vessels) influencing results
Measuring cortical thickness. Features of the cortex (thicness)
· Outer layer of grey matter
· 1-5 mm thick
· Highly folded
· 2-dimensional, embedded in 3D space
Wha is the pial surface?
Grey matter
Pros of measuring cortical thickness?
- Pros:
○ automated, continuous, whole cortex
○ processing and measurement respect cortical topology
○ direct, biologically meaningful measure in millimeters
○ surface-registration may increase sensitivity
Cons of measuring cortical thickness?
- Cons:
○ heavy post-processing (6-24 hours/scan)
○ dependent on classification
○ manual corrections often necessary
○ limited to (neo)cortex
Example of measuring cortical thickness in clinical settings
Frontotemporal dementia (FTD)
- degeneration of frontal and temporal lobes
- symptoms commonly include
- Behavioral changes
- Speech and language impediments
- In some subtypes, Parkinson-like motor symptoms
- Genetic and molecular overlap with amyotrophic lateral sclerosis (ALS)
- Currently no causal cure
Local gyrification index (GI)
Estimates the amount of cortex buried within sulcal folds, can be estimated by FreeSurfer
Regional fractal dimensionality (FD)
- Measure of ‘structural complexity’
- Quantifies how regular or irregular the shape of an object is, noting changes of brain shape overtime
- Yields a ‘broken’ topological dimension, typically between 2-3
Cortical surface model (FreeSurfer)
→ Triangle mesh (“finite element”)
→ ~ 300 000 triangles
→ Point of triangle intersections: vertex → Surface of the triangles: face
→ XYZ coordinates at each vertex
→ area, distance, curvature, thickness,
