Hello CONN-ers,

I'm attempting to use CONN for ROI-ROI and seed-voxel analyses. What are people using with regards to lesion segmentation? Are people excluding lesion voxels from the computation...additional regression...if so, where in the processing pipeline is this occurring (i.e. before or after transformation to atlas space)? Any insight would be greatly appreciated.

-KJ

Hi!
I am Brazilian medical student.
I have been trying to process some resting-state data from these patients using CONN toolbox, however, I am struggling with the fact that the lesions are segmented as grey matter. Were you able to solve this problem about the lesion segmentation? I think that maybe we should do the preprocessing steps prior on SPM using lesion masks. By the way, could you please tell us which software you use to make identify stroke lesions?

Thank you in advance for your attention!

I can't speak to the exclusions of lesions via a mask for CONN analysis, but our group has used a few packages for lesion segmentation with the SPM-based Lesion Segmentation Toolbox (LST) being one of the better packages.  No lesion segmentation approach is perfect (except for tight manual tracing), but LST does a reasonably good job and importantly for our purposes has longitudinal modules.

I would be interested to hear from others about lesion masks in the CONN processing pipeline.  Is there an option during the preprocessing set-up to allow for masks?

Warm regards,
Jeff

Hi Jeff,

Any luck solving using lesion masks in CONN? I saw elsewhere on this forum that you need to modify the TPM.nii file, but I haven't found any sources on how to do so: https://www.nitrc.org/forum/message.php?msg_id=26112 

So far, all I have found is a paper demonstrating that masking a lesion is crucial to analyzing functional connectivity in a brain with a lesion: https://journals.sagepub.com/doi/abs/10....

Thanks,

Matthew

Hi, Matthew.  Are you talking about white matter lesions or gray matter?  The aCompCor method in CONN regresses out white matter and CSF from the functional connectivity signal as part of the denoting process.  As long as white matter lesions are not being falsely segmented as gray matter, then you shouldn't have any problems.  I know that CAT12 has options to detect and either reassign white matter lesions as white matter or as its own tissue class.  In theory, though I haven't tried it yet, you should be able to import the CAT12 white matter segmentations (those with lesions assigned to white matter) into CONN at the setup stage.  This should allow any lesion-related signal to be regressed out with the normal white matter during the denoising stage.  

If, on the other hand, you are talking about gray matter lesions or encephalomalacia, then there I believe you will need to alter the analysis brain mask in the setup options.  I haven't come upon this problem yet in our patient data, so I really can't speak to how to implement that change in CONN.  Using the example above with the tissue segmentations, I guess one could assign a lesion area to the white matter or CSF tissue type mask for that person, which (I think) would exclude that lesioned gray matter area from contributing to the non-regressed signal.  

Hope this helps, though admittedly I'm a bit shaky on the correction possibilities myself (other than how we handle the white matter lesions).

Warm regards,
Jeff



Originally posted by Matthew Heard:

Hi Matthew

In the latest release there is also a (still in development) possible way to create new TPM files informed by a lesion mask. For that you may simply use the syntax:

   conn_createtpm( lesionfile, anatfile)

where 'lesionfile' is your lesion mask, and 'anatfile' is an anatomical image of the same subject and in the same space as your lesion mask. This will create a new TPM.nii file (named by default lesionTPM.nii and located in your current directory) that includes a 7th tissue class (in addition to the standard gray/white/csf/bone/soft/air categories) characterizing the lesion. You can then use this new lesionTPM.nii file during segmentation (and normalization) to make sure that the lesion is properly taken into account during the combined normalization/segmentation procedure. See "help conn_createspm" for additional details, and this is still a work in progress so if you have a chance to try this out please let me know how it goes and/or if you run into any issues

Best
Alfonso

Originally posted by Matthew Heard:

Hi Alfonso,

I'm a Brazilian Phd student and I work with Stroke patients too. I'm trying to modify the TPM file with the script that you suggested on another before I read this post. I tried to use the command conn_createtpm with different images and I found the same error message:

>> conn_createtpm
Select mask file
Select structural file
Error using conn_getinfo (line 148)
File type not implemented

Error in conn_file (line 11)
[nV,str,icon,filename]=conn_getinfo(filename,doconvert);
Error in conn_createtpm (line 50)
structural=conn_file(conn_fullfile(structural)); structural=structural{1}; 

My Best,
Alana Batista
Originally posted by Alfonso Nieto-Castanon: