Transcript fMRI QA - University of Edinburgh

fMRI Quality Assurance (QA)
(v2)
Data acquisition, processing and analysis
Katherine Lymer
Medical Physicist in MR Support
SFC Brain Imaging Research Centre for Scotland, University of Edinburgh
fMRI QA: Objectives
• Describe the quality of the fMRI data
Using measures of SNR, SFNR, fluctuation and drift
• Monitor the long-term stability of scanner performance in fMRI
• Allow between site-assessments of fMRI data
Image Acquisition
Equipment set-up
Coil: As per usual fMRI examination (e.g. quadrature / 8 channel-head coil)
Test object: Manufacturer provided homogenous sphere
Test-object set-up
lasers
asholder
with
“zero”
centre
position
of coil;
UseAlign
testMark
object
(if
applicable);
Centretest
coilobject
/ test in
object
in scanner
Place
centre
of coil;
Image Acquisition
Frequency of acquisition:
Site dependant: May be useful to acquire data on a weekly / fortnightly basis to
establish a baseline; The results from our QA suggest that a monthly test is
acceptable on our scanner.
Scanning parameters
Note magnet room temperature
Sequence: Two approaches:
(1) As per typical fMRI examination
(2) “Pushing” the scanner as hard as possible (measure stability “under stress”)
In Edinburgh, we use the following protocol on our 8 channel head coil:
TE
TR
NEX
FOV
#slices
Slice
thickness
spacing
nVols
Matrix
Flip
angle
Tscan
40ms
2044ms
1
24cm
28
5mm
0
202*
128*64
90º
6min
53sec
* Includes dummy scans (processing software expects 2 dummy vols)
Note the RF-receiver and transmit gain and the resonant frequency
Image Processing
Transfer data to “off-line” to image analysis work station
NB This processing protocol is based upon that provided by the BIRN
(http://www.nbirn.net/research/function/fbirn_tools_primer.shtm)
and is similar to that used in “Calibrain”.
See also:
Weisskoff et al, MRM 36: 643-645, 1996
Friedman et al, JMRI 23: 827–839, 2006
Requirements:
(1) PC / workstation with matlab;
(2) SPM5 (as supplied by the FIL: http://www.fil.ion.ucl.ac.uk/spm/)
(3) fMRI_qa.m script (as supplied by fBIRN:
http://www-calit2.nbirn.net/tools/fbirn_stability_phantom/index.shtm)
This script needs to be modified to reflect local / in-house needs
(4) mricro
Data Management
(1) Ensure that data is stored in central fMRI data store, in appropriate folder
denoting the current QA session
(E.g. In Edinburgh, the data is stored in Y:\MONTHLY_QA using the format:
QAddmmyyyyTEMPtt_yyyymmdd_studyNumber)
(2) In the relevant monthly QA folder, make two new sub-folders to
differenentiate between the raw and reconstructed data e.g. “raw_data”
and “SPM5_output”.
(3) Copy the fMRI dicom files into the raw_data folder
(E.g. in Edinburgh, this is:
QAddmmyyyyTEMPtt_yyyymmdd_studyNumber\raw_data)
Pre-processing
(1) Start Matlab (ensure all paths are correct)
(2) Start SPM5
(3) Convert the DICOM files using the “DICOM Import” function in SPM5:
Select the files to be reconstructed (all in the “raw data” folder)
Select “Done”;
Select output dir: SPM5_output;
Select “Done
(NB if performing reconstruction on a windows machine:
In pop-up window:
Output image format: Two file (img +hdr) NIFTI
Use ICEDims in filename: No)
This may take a few minutes
Pre-processing
(4) Start mricro
Check the central slice in volume
Load the first “real” fMRI volume (i.e. the third listed volume) into mri cro
Review entire volume and note central slice
Amend fMRI_qa.m accordingly
Image Processing: fMRI_qa_calc
Edits required to fMRI_qa (original version obtained from http://wwwcalit2.nbirn.net/tools/fbirn_stability_phantom/index.shtm)
(1) Ensure TR is correct;
(2) Ensure that the date is in the correct format;
(3) Define dimensions of ROI for ghosting measurement by changing:
X1G, X2G, Y1G, Y2G
(4) Ensure that the central ROI is placed appropriately within the central
image
(5) Check that the central slice number is correct (i.e. it matches that
practically acquired)
Image Processing: fMRI_qa_calc
Run your version of fMRI_qa
Import values into excel spreadsheet;
Plot results to help identify any trends / outliers.
Note any variation in RF-receiver / -transmit gain and resonant frequency
(since the phantom and the protocol remain the same, these should
remain constant)
Image Analysis:
All image analysis is performed on the central slice of the volume. (NB It is
important to check that the central slice as designated in KL_fMRI_qa.m
matches that practically acquired.)
Application of the BIRN / Calibrain protocol yields six measurements:
(1) Mean signal
A mean image is produced by calculating the mean, on a voxel by voxel
basis, across all the central slices.
A signal summary value is obtained from an ROI placed in the centre of this
mean image.
Example mean image:
Image Analysis:
(2) Temporal Fluctuation Noise Image
A fluctuation noise image is produced by subtracting the calculated trend
line from the data (the BIRN function uses a second order polynomial).
(It’s effectively a standard deviation image.)
By removing the trend from the data, we can estimate the fluctuations in the
data about that trend: A linear trend may indicate a systematic increase
or decrease in the data (caused by e.g. sensor drift).
Example standard deviation image:
Image Analysis:
(3) Signal-to-Fluctuation Noise Ratio (SFNR) Image and Summary SFNR
Value
A SFNR image is created by dividing the mean signal image by the temporal
fluctuation (SD) image on a voxel-by-voxel basis.
The summary SFNR value is obtained from the mean SFNR from the same
ROI placed in the centre of the SFNR image.
Example SFNR image:
Image Analysis:
(4) Static Spatial Noise
In order to measure the spatial noise, the sum of the odd and even
numbered slices are calculated separately (so there are two results); the
difference between these two sums is then calculated. This
approximates the static spatial noise.
Is there is no drift in either amplitude or geometry across the time series,
then this difference image will show no structure of the phantom and
provides a measure of intrinsic noise.
Example difference image:
Image Analysis:
(5) SNR Summary
The same ROI is placed in the centre of the static spatial noise (difference)
image.
Using the signal summary value (see (1)), the
signal summaryvalue
SNR 
((variance summaryvalue) /# timepoint s)
Image Analysis:
(6) SNR Percent Fluctuation and Drift
The average signal intensity is calculated from each of the central slices
across the time series using the same ROI, placed in the centre of each
image.
A time series of volume number vs average intensity is plotted and a second
order polynomial is fitted to the data to establish a trend line.
The mean signal intensity (prior to detrending) and the SD of the residuals
after detrending are calculated.
% fluctuation  100
% drift  100
SD of residuals
mean int ensitysignal
(max fit value)  (min fit value)
mean signal int ensitysignal
Image Analysis:
(7) Summary graphs
In addition, a series of summary graphs showing the change in raw signal,
SGR, magnitude spectrum and relative standard deviation will be
displayed
Image Analysis:
(8) Signal-to-Ghost Ratio (SGR)
In addition to the BIRN tests, a signal to ghost (SGR) measurement was
added to the protocol by Dr Gordon Waiter, Aberdeen Biomedical
Imaging Centre, University of Aberdeen.
The image may have to be windowed to observe the ghost.
In order to make the SGR measurement, a second (ghosting) ROI is defined
within the ghosting region of the image. The SGR summary value is
calculated from the mean signal (from the central ROI) / mean signal
(from the ghosting ROI).
SINAPSE Reporting Mechanism:
Short-term
Maintain results locally;
Distribute summary results with colleagues e.g. for multi-centre studies,
comparisons of scanner performance (KL happy to hold central data
base and distribute as required).
Medium - Long term
We aim to produce a SINAPSE QA reporting website to which these results
can be uploaded.
Watch this space…………..!