====== Frequently Asked Questions from the Final Report ======

This page is a reference for students working on the final project. These are the questions that we have often received this year and last during the second week of the course.

==== General ====
  * **How can I streamline my analyses so that I am running multiple subjects sequentially?**\\
      - If you create batches for an analysis step for each subject and save them, you can load them (using File --> Load batch in the Batch Window) all simultaneously. They will appear in the batch window in the order you selected them and if you press the RUN button they will run one after another.
          * To further streamline: If you named them all the same, for example: "//batch_slicetime_... .mat//", you can use the Filter function of the File selector and type in: //batch_slicetime// and then click on the button to the left with Re on it. SPM will then look in all of the subfolders for any files with that name and select them. 
      - You can set up all of the preprocessing steps or multiple statistical analysis steps for a single subject using dependencies, if you know what the output of each of the previous steps are.
      - The scripting tutorial describes how you can semi-automate the analysis by programming a bit in Matlab. It is advisable to use this step only if you have a bit of experience with programming Matlab, otherwise it might take just as long as doing it manually. 

  * **How do I know if a step of my analysis is still running?**
      * The progress bar in the lower left window in SPM usually shows you whether something is running. However, the only way to really know if your analysis step is complete is to go to the Command Window in Matlab (in the main Matlab window) and see if you have two lines of text that say "Done". It must be twice for the whole batch to be finished, otherwise just one step in your batch is complete.

==== Preprocessing ====
  * **Are there any steps from the preprocessing tutorial that I can leave out?**
      * The slice-time correction is the step that is the least common step to use. Many scientific labs do not use slice-time correction. If you do leave it out you should be able to say why. You could also theoretically leave out the detect abnormal slices step of the preprocessing, or you might conceive of a way to answer the questions for the report that does not require a group-level analysis and then you would not need to normalise the data. However, if you have non-movement related artifacts in individual subjects in your data you may not find them if you leave out the detect abnormal slices step. Also, if you choose not to do the group-level statistics you must find an alternative way of answering the questions that we posed that is at least as good as a group-level statistics. 

  * **What went wrong if my preprocessed functional images do not match the template image?**
      * Unfortunately, there are a lot of different reasons why the preprocessing could go wrong. It is extremely important that you are systematic and careful about setting up your preprocessing so that you are not the cause of the error. You could have accidentally mixed up two subjects so that you are normalizing the functional data with the deformation field of a different subject. Or you could have neglected to select ALL of the functional image frames/volumes when setting the origin. However, assuming you did everything correctly, the most error-prone step of the preprocessing is the coregistration step, which is highly dependent on how you set your origin. Therefore I can recommend that you make sure that the structural image from each subject fits well to the functional images after setting the origin. What do I mean by that? You should make sure that the origin looks like it is at the same location in the image, and that large anatomical landmarks (e.g. ventricles, or the cerebellum) are approximately in the same place. This should speed up preprocessing and eliminate the necessity to repeat steps because the normalization failed.

  * **What do I do if the preprocessed functional images do not match the template?**
      * You will need to first figure out what step did not work (typically coregistration) and then, understanding what is done in the different steps decide on what steps you will need to repeat. I typically work backwards. If the segmentation step did not work, then the anatomical image should not fit well to the template image. So I open both the template and the normalized anatomical image with Check Reg to see if the segmentation worked. If it did and if the anatomical image was the reference image then the anatomical image can be left alone. Then I would check the output images from the realignment, or the input images to the coregistration step. Check whether the functional images that were used as input to the coregistration step look like they fit well to the unchanged anatomical image. If they do not then the problem was the coregistration step and you will need to reset the origin of the functional images to match the anatomical image. You can do this with the output of the realignment step, and then you do not need to redo the realignment. 

  * **Why does the normalization step only need one functional NIFTI-file and not multiple frames from the NIFTI-file? **
      * The normalization step is part of the DARTEL toolbox which is a separate toolbox that works within SPM. This toolbox was designed to use 4D NIFTI inputs and will automatically read all frames, as opposed to SPM, which still needs you to specify each individual frame.

  * **Is it possible to load more than one image to be normalized?** 
      * Yes, you may load multiple 4D NIFTI files (for instance from two sessions of the same subject) to the images that you wish to be normalized as long as you can use the SAME deformation field for all of the images you select.

  * **When do I select "many subjects" as opposed to "few subjects" for the normalization?**
      * The "many subjects" option in the normalization step is for voxel-based morphometry, or a structural analysis method. It is use when the number of images, or frames for one subject is small compared to the number of subjects (e.g. 1 image volume per subject). In fMRI we will use the option "few subject, although we may now have more than one.

  * **Is it possible to add additional subjects to one normalization batch step?**
      * Yes, you can normalize all of the subjects (if you want the same base parameters like voxel resolution) in one batch, by adding new subjects. It is important that you add as many subjects as you have deformation fields. 

==== Single-subject analysis ====
  * **Because we want to compare the stimuli with rest, shouldn't we also explicitly model the rest periods as columns of the design matrix?**
      * It is possible but not recommended to model everything that happened in the experiment. This is because, in order to get the best estimate for the betas, or parameter estimates, the design matrix must be invertible. For a matrix to be invertible it must be full rank, full rank meaning that the number of dimensions that the matrix spans is equal to the number of columns in the matrix. If we consider the situation where we model both stimuli and the rest, (the mean is always modeled as well, see figure below) then if Stim1 + Stim2 + Rest = Mean. This means that the matrix is not full rank, and the Beta-values or parameter estimates for the individual columns may not be well estimated. It is possible that they are well estimated, but it could be that it works for 1 subject but for the rest not. {{ :course1718:designmatrix.png?600 |DesMat}}
      * However, if you use the positive contrasts (e.g. 1 0 0) for stimulus 1, you are in essence comparing them to the rest periods. 

  * **How can I look at the influence of head motion on my data?**