The volume artifact program works on both global intensity and scan-to-scan movement, so it must be run after realignment. A good opportunity is just before estimation. The input images must be resliced. The program pops up a GUI with suggested scans to be repaired, and suggested scans to be deweighted during SPM Estimation. The GUI is interactive and the user can adjust thresholds. Nothing is written until the user selects REPAIR, at which point a new set of repaired images is written. All the old images are preserved.
Start the volume detect and repair program by typing 'art_global' at the Matlab prompt. Type 'help art_global' for online help.
- How many sessions? Choose 1. RECOMMEND realign and repair each session separately.
- Program allows the first scan of each session to be dropped, if multiple sessions.
- Select the automatically generated mask
- The global mean needs a reasonable subject-specific mask. You can build one by hand and use it, but art_automask program works well (at least on the Lucas spiral data.)
- Have realignment files?. Choose Yes if you have them.
- (For raw data, realignment parameters are not available, and program is less accurate.)
- The program can show realignment parameters if they've been calculated already.
- Select image set
- The program can handle more than a 1000 scans, and from multiple folders.
- Drop first scan of each session?
- Sometimes the first volume is a warm-up, or a scanner calibration, and it should not be included in the analysis.
Let the program run until the Global Summary Figure pops up.
Global Mean and Scan-to-scan Motion Summary Figure
The summary figure has four graphs. The top graph shows the time history of the mean intensity within the head region. The time history usually has a slight downward trend, and a good run has only small bumps (less than 1.5% of the mean value). The vertical scale is the dimensionless-units that come off the scanner. The head region is defined as the voxels in the ArtifactMask which was generated during the art_global program. (Use SPM Display function to review the mask ArtifactMask.img.)
The second graph shows the size of the bumps relative to the mean of the run. The vertical scale is standard deviations away from the mean. A default threshold line is plotted at 1.5% variation from the mean. The default threshold of 1.5% from the mean is an estimate. For data from the Lucas scanner, this value attempts to keep all data within normal physiological variation and discard outliers due to sudden motion or other artifact. (You can use art_movie to inspect an outlier volume.) A good run has all its peaks below this threshold.
The third graph shows the six SPM realignment parameters. (Before realignment is done, this graph shows a crude estimate of the 3 translational estimates of centroid motion. Not to be trusted, it is just locate very peculiar scans.)
The fourth graph shows the scan-to-scan movement. The vertical scale is mm/TR. (Rotation assumes a voxel is 80 mm from the origin.) A default threshold line is plotted at 0.5 mm/TR variation. A still subject will have less than this scan-to-scan motion.
Repair and Deweight Display
The programs default thresholds are used to estimate outlier volumes and additional volumes to be deweighted during estimation.
Outlier volumes to be repaired are specified by red vertical lines. Outliers may be caused by exceeding the intensity variation threshold, or exceeding the scan-to-scan motion threshold, or by user selection. All repaired volumes are recommended for deweighting. If the data has lots of motion, we suggest using motion regressors. Then no additional deweighted scans (green lines) are necessary. If you don't wish to use motion regressors, additional volumes recommended for deweighting are specified by green vertical lines. These volumes will be close to large discontinuities in global intensity or movement EVEN AFTER REPAIR, and so may degrade the accuracy of SPM estimation.
Nothing is changed until the REPAIR button is pushed.
Threshold Adjustment Buttons
The Up and Down buttons raise and lower the thresholds (the changes are linked, no independent changes of each threshold are possible). After each threshold change, the the outliers for repair are recalculated and shown in the figure.
A rule of thumb is not to deweight more than 1/4 of the data. If the default threshold is discarding too much data, raising the threshold may improve the results.
The outlier list can be edited by hand. After Enter is pressed, the outliers for repair are recalculated and shown in the figure.
Margin Buttong For Deweighting
At any point, hitting the MARGIN button will have the program recalculate additional suggested scans for deweighting. For example, to see recommended deweighted scans, this button could be pressed after any threshold or outlier editing change.
The Repair button will repair any volumes that are outliers at the current threshold. No changes are made until the REPAIR button is hit. The volume repaired data is written with a prefix 'v'. Note that even unchanged images are copied to the new name. The repaired volumes are listed in the text file art_repaired.txt in the same folder. All volumes recommended for deweighting are listed in the text file art_deweighted.txt. (Note the volume index in these lists corresponds to order of the images...the index does not represent the volume number itself unless the first image was V001.)
We recommend using the INTERP button as the best method of Repair. This function fills values using linear interpolation from the nearest unrepaired scans. Mean fills values with the mean scan of the run. Despike fills values as a linear interpolation of the immediately preceding and following scans, whether they are repaired or not.
If repairs were done, the output files will have prefix "v" added to the name. depending on the repair method used. New files art_repaired.txt and art_deweighted.txt are written to the images folder. The repair list logs which scans were repaired. The deweighted list is for the Repair and Compare program. These files only show the sequence position of the images, so they are NOT accurate if scans are discarded as part of the estimation process.
To see the effect of repairs, you can run contrast movie or art_global again on the repaired volumes.