Question on QSDR Reconstruction and Automatic Recognition

[Manuel Leão]

Dear Frank,

I hope this message finds you well.

I need your assistance. I'm currently conducting a study with approximately 45 patients, specifically focusing on correlational tractography. 

For my metrics, I am primarily interested in FA and MD (from DTI), as they are commonly used in similar studies, and secondarily on QA and ISO (from QSDR), for novel findings :)

I have some acquisition "limitations": Non-isotropic spatial resolution (1.8 x 1.8 x 6.5); 1 low b-value (1000); insufficient diffusion sampling direction (48 directions); i dont have reverse phase encoding.

Question 1)

[Step T2 ] Pre-processing / Reconstruction

The choice of optimal pre processing parameters is tricky. I have analyzed different combinations and guided myself based on how high the R-square is. 

Is this approach correct? I feel like I might be adding bias related to R square values as the ultimate quality recontruction parameter. When in reality R is just telling us how good was the alignment/registration into template space, right?

This were the results:

[Step T2] [Image flip z] [Corrections][EDDY]

[Step T2b(1)]=QSDR

[Step T2b(1)][QSDR resolution]=1.79  (the voxel output resolution matches acquistion in-plane resolution. By default 3.0mm decreased the R value. What's the harm of using 1.79 instead of 2 mm ?) 

Question 2)

[Step C] Correlational tractography

I want to keep only "strong" findings so I used an high  T-threshold (3), as well as no FDR (unchecked) and the default length (15). The final results were interesting.

One of my goals is to specify the white matter tracts that are related to the study variable.

I have bad anatomy knowledge so I am using the Automatic Recognition [Trac Misc] [Tract Regcongition].

You have said we need to be careful interperting the output of automatic recongition, as there can be errors. But since I am using QSDR reconstructed tracts (aka in the MNI stereotaxic space), isn't it a lot more safe to interpret the automatic findings? 

Thank you in advance,

I hope I have communicated my doubts in the best way,

Best regards,

Manuel

[Frank Yeh]

The choice of optimal pre processing parameters is tricky. I have analyzed different combinations and guided myself based on how high the R-square is. 

Is this approach correct? I feel like I might be adding bias related to R square values as the ultimate quality recontruction parameter. When in reality R is just telling us how good was the alignment/registration into template space, right?

You are right. I would suggest a quality control on all SRC files. The neighboring DWI correlation and diffusion contrast (available in the most recent version) are good for identifying quality issues. This will help drop scans that have serious motion or artifacts.

 

This were the results:

[Step T2] [Image flip z] [Corrections][EDDY]

[Step T2b(1)]=QSDR

[Step T2b(1)][QSDR resolution]=1.79  (the voxel output resolution matches acquistion in-plane resolution. By default 3.0mm decreased the R value. What's the harm of using 1.79 instead of 2 mm ?) 

The slice thickness is only 6 mm here. using 1.79 is going to be a problem in registration. 2mm would be a better choice.

 

You have said we need to be careful interperting the output of automatic recongition, as there can be errors. But since I am using QSDR reconstructed tracts (aka in the MNI stereotaxic space), isn't it a lot more safe to interpret the automatic findings? 

Many pathways share a common path in some segmentation. Even if the results are in the template space, the interpretation can still be tricky.

You may post the pictures of the findings (in three views) and your anatomy labels on this forum, and I can help confirm.

Best,

Frank

 

Thank you in advance,

I hope I have communicated my doubts in the best way,

Best regards,

Manuel

--
You received this message because you are subscribed to the Google Groups "DSI Studio" group.
To unsubscribe from this group and stop receiving emails from it, send an email to dsi-studio+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/dsi-studio/a10553b3-1f56-431f-9b67-953165551221n%40googlegroups.com.

[Frank Yeh]

Sorry, I forgot to mention the Quality control. It found 1 outlier. Curiosly, out of the 5 bad slices, I could only visualize 3 of them (which I then removed using [Remove row]). 

 The neighboring DWI correlation and diffusion contrast are new to me. How do find these two? I think im using the latest version (January 4, windows)

The January version does not have the diffusion contrast.

They are at Step T1a

 

> The slice thickness is only 6 mm here. using 1.79 is going to be a problem in registration. 2mm would be a better choice.

Thank you for the advice. I will re-run the processing steps with the new output resolution. If possible, I would like to understand why 2 mm would be a better choice :)

6mm upsample to 1.79mm is going to introduce a lot fo interpolation error.

 

> Many pathways share a common path in some segmentation. You may post the pictures of the findings (in three views) and your anatomy labels on this forum, and I can help confirm.

I would appreciate that very much. Ok, here it goes (using 1.79 mm output resolution, T-score of 3, length 15  mm):

You may need to load those tracts and manually separate them into different bundles and find their anatomical names, respectively.

The tutorial videos at https://practicum.labsolver.org/ includes step to isolate tracts and identify them.

Best,

Frank