From Nipype Beginner's Guide (Some parts actually word-for-word copied; I suggest going there if you want to learn rather than have your memory jogged).
The standard anatomical volume (1mm voxels) is 256x256x256 voxels.
The scanner measures one plane (slice) of the brain after another (generally a horizontal plane). The resolution of the measured volume depends on the in-plane resolution, the number of slices/the thickness of the slices, and any possible gaps between them.
The quality of the MRI depends on:
- The resolution
- The repetition time (TR): time required to scan one volume
- The acquisition time (TA): time required to scan one slice $TA = TR - (TR/n_{slices})$
- The field of view (FOV): the extent of the slice (ex. 256mm x 256mm)
The raw data format output by MRI scanners varies with the type of scanner, and is often saved in k-space format. Common formats are DICOM and PAR/REC. I think DICOM and PAR/REC are scanner-specific formats encoded in k-space format, but I'm not sure.
I think that the k-space formatted files are converted into NIfTI or Analyze formats. These contain an image and header. For NIfTI files (extension .nii-file) the image and header are in the same file. Analyze, which is older, produces 2 files (.img for the image and .hdr-file for the header)
- High-resolution images used as reference images for
- corregistration
- normalization
- segmentation
- surface reconstruction
- Voxel resolution ranges from 0.2-1.5mm depending on the magnetic field of the scanner (1.5T, 3T, or 7T)
- Grey matter structures are dark
- White matter structures are light
Neural activity results in a characterized curve of blood oxygen level. First an initial dip in oxygen; then an increase in oxygen (peak reached after 4-6 seconds); an undershoot in oxygen levels (after 10s in figure); then eventual stabilization.
- The signal is the Blood Oxygen Level Dependent (BOLD) response
- Voxel resolution is 2-4mm depending on the magnetic field of the scanner
- Grey matter structures are light
- White matter structures are dark
Types of experimental designs:
- event-related design: Stimuli are administered for a short period. BOLD responses will be short and will manifest as peaks
- block design: Multiple stimuli of similar nature are shone in a "block" or "phase" of 10-30s. The peaks will be elevated for a longer time, creating a plateau; thus the underlying activation increase should be easier to detect.
- resting-state design: Absence of stimulation. Sometimes done to analyze functional connectivity of the brain.
Done to obtain information about white matter connections. Measures the diffusion of water via mean diffusivity (MD), fractional anisotropy (FA), and Tractography.
Includes
- diffusion tensor imaging (DTI)
- diffusion spectrum imaging (DSI)
- diffusion weighted imaging (DWI)
- diffusion functional MRI (dfMRI)
It is a new field in MRI and has problems with its sensitivity to correctly detect fiber tracts and their underlying orientation. Standard DTI has almost no chance of reliably detecting kissing (touching) or crossing fiber tracts.
High-angular-resolution diffusion imaging (HARDI) and Q-ball vector analyses were developed to account for this disadvantage.
Still from Nipype Beginner's Guide
- Template image is the standard brain in reference-space. Can be a Talairach-, MNI-, or SPM-template, or another reference image.
- Source image (normally a higher-resolution structural image) is used to calculate the transformation matrix, which is used for the rest of your images.
Used on both structural and functional data
To test a hypothesis, we need to specify a model that incorporates the hypothesis and accounts for the expected function of the BOLD signal, the movement during measurement, experiment specify parameters and other regressors and co-variants. This is usually represented by a Generalized Linear Model (GLM).
Describes a response $y$ (ex: the BOLD response in a voxel), at the time points with data in terms of all of its contributing factors ($X\beta$) and error ($\epsilon$).
$$ y = X\beta + \epsilon $$Caveats for the GLM approach:
- We need to take the time delay and the HRF (hemodynamic response function) shape of the BOLD response into account when creating the design matrix
- We need to high pass filter the data and add time regressors of 1st, 2nd, ... order to correct for low-frequency drifts in the measured data (the low-frequency noise are caused by scanner drift and other non-experimental effects)
- The high pass filter is established by setting up discrete cosine functions over the time period of data acquisition.
- Each regressors in the model decreases the degrees of freedom in statistical tests
Applying the model on the time course of each and every voxel. SPM will create images every time an analysis is performed.
- Beta images: images of estimated regression coefficients (parameter estimate). These contain information about the size of the effect of interest. A given voxel in each beta image will have value related to the size of effect for the explanatory variables.
- Error image (ResMS-image): Residual sum of squares/variance image. Measurement of within-subject error at the 1st level or between-subject error at the 2nd level analysis. Used to produce spmT images (see below).
- con images (con-images): Produced during contrast estimation from linearly combining beta images
- T images (spmT-images): Produced during contrast estimation from combining the beta values of the con-images with error values of the ResMS-image to calculate the t-value at each voxel.
1st level analysis (within-subject analysis): The data does not need to be normalized. The design matrix for this level controls for movement, respiration, heart beat, etc.
2nd level analysis (between-subject analysis): Requires normalization and transformation onto reference-space of subject specific data. Requires the contrasts of the 1st level analysis. The design matrix controls for age, gender, socio-economic parameters. Here we specify the group assignment of each subject.
Specify how to weight the different regressors of your design matrix and combine them into one image. Essentially, how can you numerically weight what you are interested in such that it's on the same scale.
Examples:
faces (+1) vs resting (-1)
session 1 (+1) vs session 2 (-1))
Specify the level of significance you test your data on. Correct for multiple comparison and specify the parameters of the result you are looking for.
Examples:
- FWE-correction: the family-wise error correction can correct for multiple comparisons
- p-value
- voxel extend: specify the minimum size of a significant cluster with the lowest number of voxels it needs to contain (ex 100)
Simple non-contrast MRI protocol - mainly for use in pain, not studies.
T1 weighted
T2 weighted
FLAIR
Diffusion weighted imaging (DWI)
Susceptibility weighted imaging (SWI)
The Brain Imaging Data Structure (BIDS defines an organizational system relying on folder and file names.
The DICOM files are organized:
dicomdir/
[timestamp]/
.dcm files
The other files follow
participants.tsv sub-01/
anat/
.nii.gz
func/
.nii.gz
.jsondwi/
.nii.gz
.json
.bval
.bvec
Functional group of different types of MRI data
sub-<participant_label>[_ses-<session_label>]_task-<task_label>[_aqc_<label>][_rec-<label>][_run-<index>]_bold.nii[.gz]sub-<participant_label>[_ses-<session_label>][_aqc_<label>][_run-<index>]_dwi.nii[.gz]sub-<participant_label>[_ses-<session_label>][_aqc_<label>][_run-<index>]_dwi.bvalsub-<participant_label>[_ses-<session_label>][_aqc_<label>][_run-<index>]_dwi.bvecsub-<participant_label>[_ses-<session_label>][_aqc_<label>][_run-<index>]_dwi.jsonsub-<participant_label>[_ses-<session_label>][_acq-<label>][_rec-<label>][_run-<index>]_<modality_label>.nii[.gz]NOTES
- Derivatives of the original data should be kept in a derivatives/ folder
- Elapsed time is in seconds
- Frequency is in Hz
- YYYY-MM-DDThh:mm:ss
FLAIR files.
It's a pulse sequence used in MRI; can be 2D or 3D. Can be used to suppress CSF effects. Helpful for studying lacunar infarction, MS plaques, Subarachnoid heamorrhage, head trauma, and meningitis. The inversion time (TI) chooses which signal is nullified.
Also in Radiopaedia. Removes signal from CSF. It is part of almost all protocols for imaging the brain. Useful in detection of subtle changes at the periphery of the hemispheres and in the periventricular region.
Quantitative T1 map files.
T1 relaxation time is the time it takes for the protons to "relax" back to their original position. It can vary for different tissue types: MS patients have widespread increases in T1 relaxation times. The map of the T1 relaxation times is the "T1map".
T1 map can have a high level of noise and poor signal to noise ratio when compared to T1-weighted image.
T1 mapping may be used to
T1 weighted image.
T1 weighted image is one of the basic pulse sequences in MRI and demonstrates differences in the T1 relaxation times of vectors.
It relies on the longitudinal relaxation of a tissue's net magnetisation vector (NMV). (How long the protons' spins realign with the main magnetic field).
It tends to have short TE (echo time - the time between the application of the radiofrequency excitation pulse and the peak of the signal induced in the coil; ms) and TR (repitition time - the time from the application of an excitation pulse to the application of the next pulse; ms) times. You need to pick a TR shorter than the tissues' recovery time to allow for contrast between the tissues.
Fat is bright; water is dark.
T1-weighted sequences include:
A defacing mask. "Mask used for defacing"
Imaging data acquired during BOLD imaging. If it is a task based fMRI, a corresponding task events file will be provided: <matches>_events.tsv. Look for a similiarly named .json file for metadata.
Deoxygenated hemoglobin is paramagnetic whereas oxygenated hemoglobin is not, and therefore the former will cause local dephasing of protons, and thus reduce the returned signal from the tissues in the immediate vicinity. Heavily T2* weighted sequences are used to detect this change, which is in the order of 1-5%.
Limitations:
T2 weighted image.
Highlights differences in the T2 relaxation time (spins decaying from the aligned precession in the transverse plane) of tissues. Requires long TE and TR times.
Each tissue has an inherent T2 value (external factors such as magnetic field inhomogeneity can increase the T2 relaxation time - these are captured in T2* weighted imaging).
Does not work well with paramagnetic contrast agents, unlike T1.
Fat is intermediate bright. Fluid is bright.
Possibly an isotropic diffusion map.
Water diffusion in the brain is hindered by cell structures, so water diffusion can occur:
The Apparent diffusion coefficient (ADC) ($mm^2/s$) is a measure of the magnitude of diffusion of water within tissue. They are calculated automatically by "the software" and then displayed as a parametric map. ADC measurements are recorded for a given region by drawing regions of interest on the ADC map.
$ ADC < 1.0*10^{-3} $ indicates restriction, but depends on the organ being imaged and the pathology.
Useful values ($10^{-6} mm^2/s$):
B value measures the degree of diffusion weighting applied, indicating the amplitude ($G$), time of applied gradients ($\delta$), and duration between the paired gradients ($\Delta$).
$$ b = \gamma^2G^2\delta^2(\Delta-\delta/3) $$The ADC is calculated using $b$ values. A good rule of thumb is to choose the $b$ value such that $b*\text{ADC} \approx 1$
Diffusion tensor imaging is an extension of diffusion weighted imaging. I'm not sure how one gets from A to B
Probably a field map.
A field map is an image of the intensity of the magnetic field across space. The magnetic field may change around air-filled sinuses and bone, which can lead to distortion of the image.
SPM provides geometric distortion correction:
Realign & Unwarp.FieldMap toolbox and VDM utility.If included with a BOLD file, according to BIDS, the BOLD file will have the metadata PhaseEncodingDirection, EffectiveEchoSpacing, TotalReadoutTime, and EchoTime.
If included with a dwi file, according to BIDS, the dwi file will have the metadata PhaseEncodingDirection, EffectiveEchoSpacing, and EchoTime.
Optionally includes an IntendedFor field with one or more relative paths stripped from sub-<label>.
A gradient echo sequence file.
In [ ]: