Our pipeline uses state-of-the-art approaches for quantifying the volume of cerebellar subunits using human structural magnetic resonance images. The process includes anatomical parcellation, voxel-based morphometry, and quality control processes. Our standardized pipeline is mostly automated, available in Docker and Singularity format, and has broad applicability to a range of neurological diseases.
To begin, ensure that Docker or Singularity, MATLAB, and SPM12 are installed. Then, using the make directory command in the command line, create folders in the working directory and label them acapulco, suit, and freesurfer. Next, in the acapulco directory, create an output folder.
In the output folder, create a directory for each subject in the study containing the t1-weighted image in the NIFTI gz format. For anatomical cerebellar parcellation, download the acapulco container, then download the relevant scripts and containers required to run acapulco. Then place the acapulco Docker or Singularity container, contents of the QCs_scripts archive, and the RCIF container or calculate_dicv.
tar file in the acapulco directory. Next, open a terminal and using Singularity, type the indicated command to run the acapulco container on a single image. Wait for five minutes for processing to complete.
Then, loop across all subjects or scans in the cohort. After processing, look for the files generated in the subject-specific folders. Identify the parcellated cerebellum mask in original and volumes for each of the 28 subunits generated by acapulco.
Then, from the pics directory, identify representative sagittal, axial, and coronal images. For statistical outlier detection and quality control, ensure that the contents of QC scripts are in the acapulco directory. Then, using Singularity, type the indicated command.
For examining the QC images generated by acapulco, open QC_Images. html in a web browser and quickly scroll through the images to identify obvious failures or systematic issues. Note the subject IDs of failed or suspect parcellated images for follow up.
Next, open Plots_for_Outliers. html to check the box plots for quantitative statistical outliers. Identify the outliers denoted by a 1 in the relevant column in the Outliers.
csv file. and note the total number of segments identified as outliers for each subject in the final column in Outliers.csv. Manually inspect each image having one or more outliers.
If using a standard NIFTI image viewer, overlay the acapulco mask onto the original t1-weighted image to check the quality of the parcellation slice by slice. If one or more parcellated regions need to be excluded from the final data set, exclude this parcellation from the analysis by replacing the volume estimate with NA in the corresponding cell of that subject. If parcellation errors result in some of the cerebellum being excluded from the mask, immediately exclude the subject from further analyses.
For voxel-based morphometry analyses using SUIT, ensure that the SPM12 folder and all sub-folders are in the MATLAB path. Additionally, ensure that the enigma_suit scripts are saved in the SPM12 toolbox directory and added to the MATLAB path. To check the MATLAB path, type pathtool in the MATLAB command window and click Add with Subfolders to add the relevant folders.
Next, using the graphical user interface, type suit_enigma_all in the MATLAB command window to run the SUIT pipeline for one or more subjects. In the first popup window, select the subject folders from the acapulco output directory to include in the analysis. Click on the individual folders on the right side of the window or right-click and select all.
Then, press Done. In the second popup window, select the SUIT directory. To call the function from the MATLAB command line for a single subject, type the indicated command.
When running SUIT from the command line, if the SPM12 or enigma_suit directories are not permanently saved to the MATLAB path, that step must be added into the command line. To call the function from the terminal window outside of MATLAB for a single subject, type the shown command. Check each subject's folder for the final outputs.
To visually inspect the normalized modulated images for major failures, type spm_display_4d in MATLAB. Next, to select the required images, navigate to the SUIT directory and type the indicated command in the filter box. Then, press the Recursive button, followed by Done.
Scroll through the images to ensure they are all well aligned. Next, to check spatial covariance for outliers, type check_spatial_cov in MATLAB. Then, select the modulated images from earlier and when prompted, set Prop.
scaling to yes, Variable to covariance out to no, Slice to 48, and Gap to 1. Finally, look at the box plot displaying the mean spatial covariance of each image relative to all others in the sample. Identify the data points that are more than two standard deviations below the mean in the MATLAB command window.
For these data points, inspect the appropriate images in the SUIT folder for artifacts, image quality issues. or pre-processing errors. Shown here is a heavily atrophied cerebellum from a spino-cerebellar ataxia 2 patient.
Despite the tissue loss seen around the edges, SUIT has warped this image to the template quite well. This would not warrant an exclusion. Here, there is a clear gradient from top to bottom of the cerebellum and the image is quite scrappy, warranting an exclusion.
Finally, in this example, the masking has not worked well. The edges are not clean and the image is smoother than those typically coming out of the SUIT pipeline. This would warrant an exclusion.
Examples of good parcellations, including healthy and heavily atrophied cerebella, are shown here, while examples of misparcellations with subtle over-and under-inclusions of individual cerebellum lobules can be seen here. These types of errors generally require the exclusion of the individual lobules that are affected, while the remainder of the parcellated cerebellum can be retained. In contrast, global failures require a complete exclusion of the subject.
When acapulco was run on a sample of 31 people with Friedreich ataxia and 37 healthy controls, the left lobule IX and right lobule Crus I had the highest exclusion rates. Comparison of volumes of 28 cerebellar anatomical lobules in Friedreich ataxia and healthy control individuals showed significantly reduced white matter in the corpus medullare in Friedreich ataxia. There were no other significant between-group differences.
Examples of well-aligned images from both healthy controls and Friedreich ataxia are shown here, while examples of exclusions can be seen here. After testing the spatial covariance of all normalized images, two scans were detected as statistical outliers based on their mean spatial covariance with the rest of the sample. Non-parametric permutation tests were carried out in SNPM to test for significant between-group differences in cerebellum gray matter volumes.
Results from the SUIT analysis showed that Friedreich ataxia patients had significantly reduced gray matter volume in bilateral anterior lobule I to V, and in medial posterior lobe regions including Vermis VI and Vermis IX.It is crucial to manually check the cerebellar masks to ensure that full cerebellar coverage has been achieved, and to also inspect the masks for over-and under-inclusions of cerebellar lobules. This pipeline facilitates multi-site group-level statistical analyses to be performed that are interested in looking at cerebellum gray matter structure. Other techniques, such as functional connectivity, can also be used to explore connectivity between individual cerebellar lobules and the cerebrum.
