Some questions regarding the analysis of MRI data

SCT
1

Hi Julien,

I hope that you are having a nice day.

I had some questions regarding the analysis of MRI data:

  1. I wanted to know that is the following list the best level choices for the cervical spinal cord (as you suggested in the analysis pipeline)?

T1: Spinal cord CSA averaged between C2 and C3.
T2: Spinal cord CSA averaged between C2 and C3.
T2s: Gray matter CSA averaged between C3 and C4.
DWI: FA in WM averaged between C2 and C5.
MTS: MTR in WM averaged between C2 and C5. Uses MTon_MTS and MToff_MTS.
MTS: MTSat & T1 map in WM averaged between C2 and C5. Uses MTon_MTS, MToff_MTS and T1w_MTS

If yes, and I want to analyze all levels, is it possible that resultant measurements and segmentations are not precise and reliable? If we’re going to change the above-suggested levels (to include more levels), what would be another choice?

  1. In terms of the resolution of the MRI data, what is your suggested MRI modality? (to get better segmentation and shape analysis results)? Is there any reference paper that explains this question (I did not find any)?

  2. Also, I want to do analysis using SCT faster; it takes so much time to run and reach the results for just one image. Do you have any suggestions to make the process faster?

Thanks very much, looking forward to hearing from you.
Maryam

2

Hello Maryam,

I wanted to know that is the following list the best level choices for the cervical spinal cord (as you suggested in the analysis pipeline)?
T1: Spinal cord CSA averaged between C2 and C3.
T2: Spinal cord CSA averaged between C2 and C3.
T2s: Gray matter CSA averaged between C3 and C4.
DWI: FA in WM averaged between C2 and C5.
MTS: MTR in WM averaged between C2 and C5. Uses MTon_MTS and MToff_MTS.
MTS: MTSat & T1 map in WM averaged between C2 and C5. Uses MTon_MTS, MToff_MTS and T1w_MTS

This was only a suggestion based on a couple of elements (eg: C2-C3 is traditionally used for atrophy measure in MS, C2-C5 has a reasonable coverage and avoids lower cervical cord data where quality is sometimes questionable, etc.), but again, this is a “general” recommendation and you should adapt it based on your data/hypotheses.

If yes, and I want to analyze all levels, is it possible that resultant measurements and segmentations are not precise and reliable?

it depends on your data and registration pipeline.

If we’re going to change the above-suggested levels (to include more levels), what would be another choice?

You can select any levels you want, as long as it makes sense from a statistical power standpoint, data quality, etc.

In terms of the resolution of the MRI data, what is your suggested MRI modality? (to get better segmentation and shape analysis results)? Is there any reference paper that explains this question (I did not find any)?

I like the T2w. See this recent publication for further justifications.

Also, I want to do analysis using SCT faster; it takes so much time to run and reach the results for just one image. Do you have any suggestions to make the process faster?

You can parallelize processing across CPU cores with sct_run_batch. See more details in this publication, where 260 participants were processed in ~40 minutes.

3

Thanks very much, Julien. I just have one concern, about the T2s, because it usually seems to have a higher resolution when I look at it.
And my last question: is there any specific reference/paper/book that I can refer to when it comes to choosing different MR modalities for different image analysis tasks?
P.S. I tried parallel processing it did not decrease the processing time very much, maybe it is because of my system limited features… I’m not sure.

Thanks and regards,
Maryam

4

Thanks very much, Julien. I just have one concern, about the T2s, because it usually seems to have a higher resolution when I look at it.

Why having an image with high resolution is a concern?

And my last question: is there any specific reference/paper/book that I can refer to when it comes to choosing different MR modalities for different image analysis tasks?

I recommend going through SCT’s tutorial, which lists the majority of quantitative analysis with a multimodal dataset.

P.S. I tried parallel processing it did not decrease the processing time very much, maybe it is because of my system limited features… I’m not sure.

It could be RAM and/or CPU limitation. You need to look into what is limiting you and adjust the parameters accordingly.

5

Because when I want to choose one modality to annotate and make ground truth, the higher the resolution and the clearer the cord regions, would works better, and leads to more accurate results as well. However, to do analysis using SCT, it is not of much concern, by doing registration first and then segmentation.