🟢 QA and MRIQC: Practice 3 (~20-minutes)

By Dianne K Patterson and Manojkumar Saranathan at the University of Arizona

Practice 3: Review Processing Results (~20-minutes)

Processing Results

If MRIQC ran correctly at the participant level, you should see two subdirectories (anat and func) in derivatives/mriqc/sub-CAM003/ses-01brain and four participant-level HTML reports in derivatives/mriqc.
If MRIQC ran correctly at the group level, you should also see:
group_bold.html group_bold.tsv group_T1w.html group_T1w.tsv, and group_T2w.html group_T2w.tsv
in derivatives/mriqc.
The logs directory will contain information only if there was a crash.
The anat and func directories contains JSON files holding the results of running MRIQC (in other words, these are NOT copies of the BIDS JSON files).
✅ Start an Interactive desktop on the HPC to view the HTML reports with Firefox.

View a Group Structural HTML Report

To view the content of an HTML report, you need to open it in a web browser. You can either download the derivatives directory to your computer or start an interactive desktop on the HPC and use Firefox. Three group-level HTML reports are in derivatives/mriqc. All are labeled group*.

View a Group Report

✅ View group_T1w.html. From the terminal, navigate to the derivatives/mriqc directory. On the Interactive desktop, you can invoke Firefox from the terminal, e.g.,

Alternatively, you can navigate to the directory in the file explorer, and then click on an HTML report to open it in Firefox. Either way, you see a very sparse quality report for one participant.

View an Individual Structural Report

Clicking a dot in one of the graphs takes you to the Individual T1w report sub-CAM003_ses-01brain_T1w.html.
✅ On the upper left, click (blue underline) each of the three sections: Summary Visual reports Other
✅ Download and examine the IQM cheatsheet, if you have not already.
✅ Select the Rating widget (upper right, green underline). You can submit a report on the quality of the subject's image. To submit your ratings, you should develop a better understanding of how IQMs relate to subjective evaluations.

MRIQC individual report

By Dianne K Patterson at the University of Arizona

Was this helpful?

(Voting helps us personalize your learning experience!)

Background Noise

✅ Scroll through the report.
✅ The Zoomed-in (brain mask) section is a lightbox view of the slices. This will allow you to identify bad scans (movement or artifacts)
✅ The colored Background noise section is especially useful to identify ghosting, eye spillage, or movement.
Most data displays these types of noise to some extent. This dataset is perfectly reasonable despite not being perfect.

Spatial Normalization: Is it Good Enough?

✅ Scroll to the Approximate spatial normalization section.
✅ In the upper left corner is a label called moving. Click it to see the quality of registration. The fixed image is the MNI template and the moving image is sub-CAM003 registered into MNI space. As expected, the structures are roughly aligned:
✅ Look at the ventricles in the axial slices.
✅ Look at the corpus callosum, brainstem, and cerebellum in the sagittal views.
This is pretty typical of normalization and looks fine. Generally, when normalization goes bad, it goes REALLY bad (e.g., the brain is upside-down, squished, stretched, or stuck in one corner of the image). That is all you are looking for, some kind of shockingly bad alignment.

Note

The participant's images are messier looking because the brain has not been perfectly extracted. There is non-brain tissue around the outside of the brain. This is all perfectly normal. Scroll down to the sections **Brain mask** and **Brain tissue segmentation**. The algorithm has done a good job of identifying the brain and distinguishing tissue types.

✅ Scroll through the rest of the sections, again, nothing is shockingly bad.
✅ Select each section under Other.

Extracted Quality Metrics contains the actual values of each image quality metric, but these are also available as text, which is more useful for analysis.

group_T1w.tsv

✅ View group_T1w.tsv. It contains the Extracted Quality Metrics in a spreadsheet, one image per row.

Note

The values are separated by tabs (tsv=tab-separated values) and can be displayed in Excel, though you may need to import the file and tell Excel that tabs are used to separate the values.

Group bold.html report

✅ View group_bold.html (bold is fMRI).
There was only one bold run in this dataset, but if there had been multiple bold runs, there would have been a report for each one.
✅ The group_bold.tsv contains the relevant measures for this fMRI run. There is one row corresponding to the single fMRI sequence.
✅ Open the individual bold report by clicking on a dot in a plot.

Individual bold report

✅ Evaluate the Standard deviation map section. You see yellow around the edges of the head, in the eyeballs, near the brainstem, and 4th ventricle. This is pretty normal and indicates movement…mostly because the participant is alive.
✅ Scroll to the FMRI summary plot section. This is a very informative section of the report.
✅ The carpet plot is now called the Royal carpet plot. Have a look at the following article to understand why (and to answer a question below): Provins C, Markiewicz C, Ciric R, Goncalves M, Caballera-Gaudes C, Poldrak R et al. (2021) Quality control and nuisance regression of fMRI, looking out where signal should not be found. Neuroimage Preprint: 2142-2154.

Individual bold report continued

✅ Look at the ICA components. They are labeled "C1", "C2" etc. The component numbers may be different every time you run MRIQC (this is true of most ICA algorithms). In blue is a graph of the time course. In orange is a graph of the frequency components. In general, noise components are characterized by ringing around the head, signal in the ventricles, and signals with a single spike. This is a resting-state scan, so there are no task-related patterns to search for.
✅ Look at the Approximate spatial normalization section. Just like for the structural image, click it to see the quality of normalization. The image looks pretty well aligned.
✅ Point your cursor at a feature (like the top of the cerebellum, or a ventricle to see how well it is aligned with the MNI template. Again, this is a pretty good job.

Digest

Summary

You have now had practice with running MRIQC: a BIDS app to perform quality assessment in a standard way.
You performed both participant-level and group-level processing. These two levels are often (but not always) available in BIDS apps, and they have distinct outputs.
You should expect that MRIQC will be an important part of your process going forward. Every participant should pass through a quality assessment and you should review the resulting HTML reports for gross problems. There is no point in further processing of data that does not meet your quality standards.
Along the way, you've gotten more experience editing, running, and troubleshooting SLURM processing scripts, and you have gotten more opportunities to use the interactive desktop on the HPC.