Quality control on Qrater

By Etienne Aumont with input from Gleb Bezgin & Vladimir Fonov

1. Accessing Qrater
2. Using Qrater
3. MRI non-linear segmentation QC
4. PET native space alignment QC

1. Accessing Qrater

Vlad has set up Qrater software for web-browser quality control (QC). The software itself is described here.

We have 3 options to access Qrater:

1.1 Using the lab's computer

In the terminal, simply enter (using the information from Vlad's email) :

 ssh IP.sent.by.email -L number:IP.2.from.email:number 

1.2 On your own computer using the McGill network (or VPN)

First, connect to the McGill WPA WIFI OR connect to the McGill VPN like for X2Go.
Second, open a terminal (search "cmd" in the Windows search bar)
Third, enter (using the information from Vlad's email):

 ssh username@IP.sent.by.email -L number:IP.2.from.email:number 

Make sure you use the lab's username

1.3 Connecting to the BIC server (requires to request an account)

First, open a terminal (search "cmd" in the Windows search bar)
Second, enter (using the information from Vlad's email):

 ssh username@IP.sent.by.email -L number:IP.2.from.email:number -J bicusername@login.bic.mni.mcgill.ca

No matter the method to connect, you can next go to your web browser and go to Qrater with this information:

 http://IP.2.from.email:number/

2. Using Qrater

Once on Qrater, things should look like this:

A bunch of image blocks with a couple of links. **Image blocks will be allocated to you - your name will be at the end of their titles. ** Before doing anything, you'll need to create an account (top right corner)
This account is independent from the BIC or the lab account.
When connected, Qrater automatically shows you only what you've done. Click on "All Raters" to see what the whole team did so far.

Regarding the different buttons for a given block:

• Images shows you the files list of all images
• Ratings shows you the files list of all images already QCed
• To see all of the images themselves, click on "rate". "pass" or "Pending" will only show the images that need to be QCed or have been already.

Images look like this:

The red line represents the target image (MRI template if it's an MRI, or subject's MRI if it's a PET). The image should roughly match the red outline. Note that the outline isn't perfect, and meningeal tissue might be segmented. We'll see examples of that further down.

I recommend using the keybound options to QC:

• "1" for pass, "2" for unsure, and "3" for fail
• Left and right arrows to go to the previous or the next image

3. MRI non-linear segmentation QC

In this type of QC, we look at subjects' MRIs that have been warped to the standard template. They are generally not perfect, but they should match the contours of the template brain (in red).

Good examples:

Bad examples:

Here, the dorsal brain apex is too low compared to the template, which means that cortical signal in standard space might actually be from this subject's meninges!

This one is ambiguous because of the orbital bone going inside the red line second image from second row. The extreme left hippocampal atrophy is not great either.

4. PET native space alignment QC

In this type of QC, PET images have been aligned with the corresponding MRI (red outline). This means that the PET hasn't been deformed. There is a large degree of heterogeneity between PET image types and subjects. Sometimes, the red line will be confusing since it will show the outline of the meninges instead of just the brain (the next image shows it well).

Amyloid-PET

Good Amyloid-PET positive images are very easy to verify thanks to good contrasts between grey matter and CSF. However, beware of the meninges' contour circled in the screenshot. The inner red line follows the cortex, but the outer one is meningeal.

Amyloid-PET negative images don't have a good signal in grey matter, so they are harder to judge.

Bad Ventricules don't match, nor does the occipital cortex. Maybe the wrong MRI was used.

Tau-PET

MK negative images usually have strong meningeal binding, making it easier to align with the outside of the brain.

Sometimes, MK has lower meningeal binding, and the assessment is more difficult due to low contrast. There is more guesswork, but this one looks good.

Tau-PET positive images are easier to judge, although the signal will be more localized than amyloid-PET.

This one can be thought as bad because of the supraorbital binding that might be interpreted as cortical, but that's actually bone binding. It's a good registration.