1
Summary Advanced
neuroimaging 2022
Evelyne Fraats
Content
MRI physics....................................................................................................................2
Quantitative MRI............................................................................................................3
Segmentation Parcellation.............................................................................................4
Registration....................................................................................................................6
Atlasing..........................................................................................................................8
Cortical and laminar analysis.......................................................................................10
Diffusion MRI................................................................................................................12
Resting state MRI.........................................................................................................14
Multimodal MRI approaches.........................................................................................16
, 2
MRI physics
The MRI scanner consists of a strong magnet (induce magnetic field – no metal!),
gradient coils (for GRE, DWI and slice selection) and an RF coil (transmit and receive
radio waves).
Resolution of the scanner; the higher the structure and better definition and faster
acquisition and quantitative MRI possible. However, also more distortions, signal
inhomogeneities and high-resolution scanners are expensive.
In the human body are lots of hydrogen atoms that are always spinning. When placed
in a magnetic field they align and process (spin around the axis) with a certain
frequency related to the strength of the magnetic field (Larmor Frequency). When an
RF pulse is applied protons flip from the z-plane (longitudinal plane) to the x-y plane
(transverse plane) and become aligned. Therefore, a signal can be measured using
the inducing properties of a coil. The two planes have different relaxation properties.
The longitudinal relaxation (so the protons coming back to the z-axis) is called T1
relaxation (if short = high signal). The transversal relaxation (protons losing
coherence in the XY plane and coming back to the z plane) is called T2 relaxation; a
long T2 means a high signal. Specific materials in the body have different T1 and T2
times. To investigate different contrast, specific TR (repletion time) and TE (echo
time, time of signal measuring) can be combined to maximize a certain desired
contrast. However, the measured signal is always a combination of both T1 and T2
relaxation.
There are several echo measures functions. Spin echo SE first applies a 90 degrees
pulse to realign the protons in the XY plane. The protons start to dephase. Then a
180 degrees pulse is applied to flip the protons, so dephasing happens in the other
direction and a reliable average can be obtained. There are some types of spin echo
sequences e.g., using a convolution inversion recovery, in which you null the signal
for a specific tissue type and thereby remove this tissue type influence on the signal
(STIR – fat, FLAIR – CSF). Another way of measuring the signal is GRE, where you
explicitly dephase and rephase spins over time. This results in a quicker signal than a
spin echo. However, in GRE you always have T2* inhomogeneitibilities effects. An
over-coupling method is EPI (can be both SE and GRE), in which you read one k-space
plane/slice in one go by applying a series of excitations (however, less resolution and
more magnetic susceptibility effects).
When the signal is measured, there is no spatial specificity label, therefore the
gradient coils are linearly changed across the magnetic field. If the magnetic field
changes, so are the Larmor frequency. This gradient coil is changed in the x, y and z
planes. The signal is saved in a k-space, in which each point reflects the phase and
magnitude of a certain spatial frequency. The lower spatial resolution is saved in the
centre and contains contrast information. The higher spatial resolutions are saved in
the boundaries and contain tissue boundaries. A combination of all these spatial
frequencies can be used to construct an image. An inverse Fourier transform is used
to go from frequency to time domain.
, 3
Quantitative MRI
In qualitative MRI, as described in the previous chapter, the signal is always weighted
across multiple contrasts (e.g., both T1 and T2). Using quantitative MRI, you can get
an unweighted image and the actual quantitative numbers of different tissues. Those
are quietly related to biophysical contrast such as myelin and iron. Quantitative MRI
combines several contrasts to model the unweighted contrast. A con of this method
is there is tissue-specific noise which is less homogenous, orientation sensitivity and
tissue-specific variability.
T2 weighted to T2. Acquire multiple images with multiple echoes (TE). An
exponential function can be made of the Signal depending on the S0 (constant value)
and TE/T2*. If you log this exponential function, it becomes easier to work with and
becomes a linear problem. You can do linear regression and use OLS as a minimizing
cost function (minimize the distance between y real and y estimate). You can take
the derivative of this function which will be 0 at the maximum. You do this for both
finding T2* and S0. If the sample ratio is uneven or if there is a very slow decay or
error/noise normality problem, another form of OLS might be appropriate.
From T1 weighted to T1. Use different TR, however in the Mp2RAGE sequence
only two inversion times (two different TR times). Create a bias-corrected T1
weighted image and look-up from stimulated T1W image the corresponding T1 value.
Different lookup tables for certain flip angles or certain ROI.
From T2 to QSM. QSM = measure of how much a material will become magnetic in
an applied magnetic field. Not only signal magnitude is important but also the raw
phase images are used. QSM calculations are quite complex.
Summary Advanced
neuroimaging 2022
Evelyne Fraats
Content
MRI physics....................................................................................................................2
Quantitative MRI............................................................................................................3
Segmentation Parcellation.............................................................................................4
Registration....................................................................................................................6
Atlasing..........................................................................................................................8
Cortical and laminar analysis.......................................................................................10
Diffusion MRI................................................................................................................12
Resting state MRI.........................................................................................................14
Multimodal MRI approaches.........................................................................................16
, 2
MRI physics
The MRI scanner consists of a strong magnet (induce magnetic field – no metal!),
gradient coils (for GRE, DWI and slice selection) and an RF coil (transmit and receive
radio waves).
Resolution of the scanner; the higher the structure and better definition and faster
acquisition and quantitative MRI possible. However, also more distortions, signal
inhomogeneities and high-resolution scanners are expensive.
In the human body are lots of hydrogen atoms that are always spinning. When placed
in a magnetic field they align and process (spin around the axis) with a certain
frequency related to the strength of the magnetic field (Larmor Frequency). When an
RF pulse is applied protons flip from the z-plane (longitudinal plane) to the x-y plane
(transverse plane) and become aligned. Therefore, a signal can be measured using
the inducing properties of a coil. The two planes have different relaxation properties.
The longitudinal relaxation (so the protons coming back to the z-axis) is called T1
relaxation (if short = high signal). The transversal relaxation (protons losing
coherence in the XY plane and coming back to the z plane) is called T2 relaxation; a
long T2 means a high signal. Specific materials in the body have different T1 and T2
times. To investigate different contrast, specific TR (repletion time) and TE (echo
time, time of signal measuring) can be combined to maximize a certain desired
contrast. However, the measured signal is always a combination of both T1 and T2
relaxation.
There are several echo measures functions. Spin echo SE first applies a 90 degrees
pulse to realign the protons in the XY plane. The protons start to dephase. Then a
180 degrees pulse is applied to flip the protons, so dephasing happens in the other
direction and a reliable average can be obtained. There are some types of spin echo
sequences e.g., using a convolution inversion recovery, in which you null the signal
for a specific tissue type and thereby remove this tissue type influence on the signal
(STIR – fat, FLAIR – CSF). Another way of measuring the signal is GRE, where you
explicitly dephase and rephase spins over time. This results in a quicker signal than a
spin echo. However, in GRE you always have T2* inhomogeneitibilities effects. An
over-coupling method is EPI (can be both SE and GRE), in which you read one k-space
plane/slice in one go by applying a series of excitations (however, less resolution and
more magnetic susceptibility effects).
When the signal is measured, there is no spatial specificity label, therefore the
gradient coils are linearly changed across the magnetic field. If the magnetic field
changes, so are the Larmor frequency. This gradient coil is changed in the x, y and z
planes. The signal is saved in a k-space, in which each point reflects the phase and
magnitude of a certain spatial frequency. The lower spatial resolution is saved in the
centre and contains contrast information. The higher spatial resolutions are saved in
the boundaries and contain tissue boundaries. A combination of all these spatial
frequencies can be used to construct an image. An inverse Fourier transform is used
to go from frequency to time domain.
, 3
Quantitative MRI
In qualitative MRI, as described in the previous chapter, the signal is always weighted
across multiple contrasts (e.g., both T1 and T2). Using quantitative MRI, you can get
an unweighted image and the actual quantitative numbers of different tissues. Those
are quietly related to biophysical contrast such as myelin and iron. Quantitative MRI
combines several contrasts to model the unweighted contrast. A con of this method
is there is tissue-specific noise which is less homogenous, orientation sensitivity and
tissue-specific variability.
T2 weighted to T2. Acquire multiple images with multiple echoes (TE). An
exponential function can be made of the Signal depending on the S0 (constant value)
and TE/T2*. If you log this exponential function, it becomes easier to work with and
becomes a linear problem. You can do linear regression and use OLS as a minimizing
cost function (minimize the distance between y real and y estimate). You can take
the derivative of this function which will be 0 at the maximum. You do this for both
finding T2* and S0. If the sample ratio is uneven or if there is a very slow decay or
error/noise normality problem, another form of OLS might be appropriate.
From T1 weighted to T1. Use different TR, however in the Mp2RAGE sequence
only two inversion times (two different TR times). Create a bias-corrected T1
weighted image and look-up from stimulated T1W image the corresponding T1 value.
Different lookup tables for certain flip angles or certain ROI.
From T2 to QSM. QSM = measure of how much a material will become magnetic in
an applied magnetic field. Not only signal magnitude is important but also the raw
phase images are used. QSM calculations are quite complex.
