Prototype of processing pipeline

[jcohenadad]

Let's get started with the dummy dataset #1

subsequent issues:

• - segment lesion
• - normalize CSA with PAM50 (see work of Sandrine and Jan)
• - replace vertebral labeling with TotalSpineSeg
• - compute lesion occupancy within lateral CST
• - define strategy to compute local atrophy at lesion levels (consider other metrics than CSA)

[plbenveniste]

I investigated the detection of local atrophy in the spinal cord on the data that we have received. I was planning on using the following method:

• if locally the CSA or the R-L diameter is less than 80% of the above and lower vertebral level, then there is local atrophy.

However, as we can see on the figure below, because there is quite a lot of variability of both the CSA and the R-L diameter, this method wouldn't work.

Moving on, atrophy should be detected by measuring deviance between the subject CSA and/or R-L diameter and average CSA or R-L diameter of a population.
Some questions emerge from this:

• Do we have a template/atlas of the evolution of the CSA and/or R-L diameter?
• If not, we should build one. Should we use the data given in this project in order to work on the same contrast ?
• Should we differentiate men and women ?

Maybe these questions could be raised during next meeting.

[jcohenadad]

Moving on, atrophy should be detected by measuring deviance between the subject CSA and/or R-L diameter and average CSA or R-L diameter of a population.
Some questions emerge from this:
Do we have a template/atlas of the evolution of the CSA and/or R-L diameter?

Hence my comment above "see work of Sandrine and Jan" 😉 have a chat with @valosekj @sandrinebedard, they can show you how

EDIT 20241125_211019: Ah! It seems like you are already aware:

critical-ms-lesion/batch_processing.sh

Line 168 in 028b073

-perslice 1 -o "${PATH_RESULTS}"/"${SUBJECT}"_CSA.csv -append 1 -normalize-PAM50 1

[valosekj]

Nice figures, @plbenveniste. Just fyi, if you need to show both PAM50 slices and vert levels on a single x-axis, you could use this script, or its parts. It generates the following figure:

The figure shows single-subject morphometrics for two sessions. Together with normative values computed from spine-generic (see this repo for one CSV file for each subject from the spine-generic dataset). The normative values can be filtered by sex.

---

what is the ref for the method you mentioned?

if locally the CSA or the R-L diameter is less than 80% of the above and lower vertebral level, then there is local atrophy.

[plbenveniste]

Thanks for your comment @valosekj. If I understand correctly, you don't have an average CSV file of the entire cohort. You only have one CSV file per subject.
So if I were to use your data, I would have to get all CSV files, average the CSA or R-L diameter per sex so that I could use it as a standard reference. Is that correct ?

what is the ref for the method you mentioned?

No method in particular, just my idea of how to solve this. It proved to be a failure.

[sandrinebedard]

So if I were to use your data, I would have to get all CSV files, average the CSA or R-L diameter per sex so that I could use it as a standard reference. Is that correct ?

yes, we provide it per subject, to allow custom filtering (like per sex),

this is what we use in sct_compute_compression -normalize-hc 1 (https://spinalcordtoolbox.com/stable/user_section/command-line/sct_compute_compression.html)

[valosekj]

So if I were to use your data, I would have to get all CSV files,

"all CSV files" from the PAM50-normalized-metrics are automatically downloaded during SCT installation (to $SCT_DIR/data/PAM50_normalized_metrics folder) because we use them for sct_compute_compression as mentioned by Sandrine.
You can load all the CSV files into a single dataframe using this function.

average the CSA or R-L diameter per sex so that I could use it as a standard reference. Is that correct ?

Once you load all the CSV files into a single dataframe, you can plot them as mean and std directly using sns.lineplot errorbar='sd', see this line. i.e., no need to average them manually