SUGGESTED PROTOCOL for ARTIFACT DETECTION AND REPAIR


There are three areas to review and possibly repair your data: 

  1. Before any SPM preprocessing, 

  2. After preprocessing and before model design, and 

  3. After statistics are calculated. 

Each of these sections is summarized below.



1.  Before SPM preprocessing ...

  The objective in this section is to detect and repair outliers in the data.

  GLOBAL VARIATE
  Run GlobalVariate (the artdetect5 program) with the automatically generated mask
  and without realignment parameters. The default threshold is 1.5% from the mean.
  Repair data outside this threshold. If the subject has very clean data, where
  all the data is below the default threshold, consider lowering the threshold to
  4 standard deviations. (The vertical scale on the second graph of the
  output is in standard deviations. The size of the standard deviation is listed
  at the bottom.)

  Repair the data with Interp option, if all outliers are isolated volumes.
  Repair the data with Mean option, if some outliers are consecutive volumes.

  The volume repaired data is written with a prefix �v�. The outlier volumes are
  noted as �outlier� images in the directory. It is best to omit the outlier scans
  during model estimation, but results will be more accurate even if the 
  repaired scans are kept in the model.
       
  SLICE REPAIR     
  Run SliceRepair (the spike_repair program) with the Repair Bad Slices option. 
  This run will write a logfile showing all the bad slices detected. The slice 
  repaired data is written with a prefix "g". If the logfile shows clean data, 
  there were no bad slices detected at the default threshold.  

  If TR<= 2 sec, you have the option of median filtering the data. This filter is
  a temporal filter for outliers, and it often helps clean up the data. It is 
  especially good if bad slices were detected, because there are usually 
  moderately bad slices just below the threshold. In this case, delete the "g" 
  files from the previous step, and rerun with the median filter option. A prefix
  "f" will be prepended to the name.

  CONTRAST MOVIE
  It's always good to know what is in your data! 
  Use the biac_movie program to run the raw data in movie mode to 
  review all the data and look for unusual patterns or data fluctuations. Run the
  repaired data to see if the repairs reduced or removed the fluctuations. If 
  desired, run the data in slider mode to inspect the data in individual volumes. 
  Note that the Tools-Zoom-In button in the image toolbar allows inspection of 
  individual voxels.

  GLOBAL AND MOTION OUTLIERS
  This is a tool for automatic and manual detection of outliers.
  This utility asks for image files and a text motion parameters file.
  It then displays four graphs:
  - The top graph is the global brain activation mean as a function of time.
  - The second is a z-normalized (stdv away from mean) global brain activation
    as a function of time.
  - The Third shows the linear motion parameters (X,Y,Z) in mm as a function of time
  - The fourth shows the rotational motion parameters (roll,pitch,yaw) in radians as
    a function of time.
  using default threshold values for each of the bottom three graphs we define outliers 
  as points that exceed the threshold in at least one of these graphs. The thresholds 
  are shown as horizontal black lines in each of the graphs.
  Points which are identified as outliers, are indicated by vertical red lines in the 
  graphs that correspond to the outlying parameters. For example, the if the absolute value 
  of the Y motion parameter for time t=17 is above the motion threshold, it is identified 
  as an outlier and indicated by a red vertical line at t=17 in the third graph. 
  The union of all outliers is indicated by vertical lines on the top graph. 
  The list of outliers is also displayed in the editable text box below the graphs.
  The current values of the thresholds are displayed by the side of the corresponding 
  graphs. These values may can be changed by the user either by pressing the up/down buttons, 
  which increment/decrement the current value by 10%, or by specifying a new value in the 
  text box.
  In Addition, the user can manually add or remove points from the list of outliers by editting
  the list. Note that the list is only updated once the curser points outside the text box 
  (i.e. click the mouse somewhere outside the text box). Since any changes made by the user are
  overridden once the thresholds are updated, it is recommended to do any manual changes as the
  last step before saving.
  Pressing the save button lets the user choose wheter to save the motion statistics (.mat or .txt)
  the list of outliers (.mat or .txt), or save the graphs (.jpg, .eps or matlab .fig).
  Please report any bugs or needed improvements to Shay Mozes shaym@mit.edu



  OUTPUT FILES
  If no repairs were needed, the original scans go on to SPM preprocessing. If 
  repairs were done, the output files may have prefixes of "v", "f", and/or "g" 
  depending on the repair method used.



2.  Before model design, and after preprocessing ...


  These tests will check normalization, total motion, and correlations between the
  task and motion. The data is assumed to be preprocessed in SPM up through
  normalization.

  CHECK NORMALIZATION
  Run SPM Check Reg to verify that normalization worked properly. Display a 
  normalized functional and anatomical image. Be sure they are not clipped by the
  bounding box. If they are, use Defaults in SPM to change Spatial Normalization, 
  and reset the bounding box to a larger size.

  CHECK TOTAL MOTION
  Run the MovementParameters program. This program displays the realignment
  parameters found by SPM preprocessing. If all motion is less than 3 mm, the data
  is OK for total motion. If some volumes exceed these bounds, then note their
  numbers and omit them from the analysis.

  ( Currently in development...
  If some data exceeds 3 mm, run GlobalVariate on the normalized data using 
  automatically generated mask, and specify the realignment parameter file
  calculated by SPM. Use the motion threshold to add those volumes to the outlier
  list. This program will also output the mean global signal to use as a possible
  covariate in the design matrix.)

  TASK CORRELATION 
  Run the Task Correlation ( the rd_taskcorr2 program). Check correlations between
  motion and task stimuli. If no task correlation and there is over 2 mm of
  motion, it�s good to add motion covariate explanatory models to the design
  matrix to reduce false activations. If motion is highly correlated with the
  task, don't add the motion covariates because they will degrade the accuracy
  of the results. 

  SPECIFY THE OMITTED SCANS
  If there were outlier volumes (from global intensity or total motion), omit them 
  when requested in SPM Estimate Model. Use the index numbers from artdetect5, not 
  the volume numbers themselves.



3. After statistical estimation ...

  The objectives are to check the geometric validity of results.

  MASK IMAGE ( GLMMASK )
  Check the mask image (Mask.img) for holes using DisplaySlices in the Toolbox,
  or Display in SPM. If there are holes in the SPM-generated mask, you'll need to 
  generate an explicit mask, and then redo the analysis using the GlmMask2
  program. GlmMask will ask for the new mask, and then rerun all the SPM results.

  An explicit mask for full-head scans can be generated automatically by running 
  GlobalVariate on a few "normalized" images- this program will generate a mask
  called ArtifactMask in the image folder. (There may be an old file with the same
  name which was used as an artifact detection mask for the un-normalized images. 
  It can be deleted.) Check the automask quality, in the same way as for Mask.img.

  CHECK ANATOMICAL REGISTRATION
  Run Check Reg in SPM on the ResMS.img, RPV.img, Mask.img to look for artifacts. 
  Run Check Reg in SPM on the contrast and beta images to look for artifacts. 
  Normal activation patterns should lie inside the head region, and be active in
  the regions that would be expected for the experiment.

  TROUBLESHOOTING
  Types of artifacts that might be seen include abnormal shapes such as lines, 
  streaks, and rings, or total blotchiness over the image, or in wrong areas such
  as outside the head or in the ventricles, or perhaps no activations at all. In
  these cases, you'll need to revisit the artifact repairs. You can review the
  processed data that enters the model by running biac_movie on the normalized
  scans. If this data looks bad, there was probably a preprocessing error. If
  this data looks good, there may be a model error.
 





