shapeworks-users

Text archives Help


Re: [shapeworks-users] Questions about ShapeWorks


Chronological Thread 
  • From: Shireen Elhabian <shireen@sci.utah.edu>
  • To: quentin devignes <qdevignes@gmail.com>
  • Cc: shapeworks-users@sci.utah.edu, Shireen Elhabian <shireen@sci.utah.edu>
  • Subject: Re: [shapeworks-users] Questions about ShapeWorks
  • Date: Fri, 31 Jan 2020 07:31:39 -0700

Hi Quentin,

Happy new year! 

- We carried out all the steps on our groups one after the other (we have 4 groups in our study). However, I wonder if a mean shape on all patients is needed for the following analyses (that means we should carry out all the steps on our entire population) 

Yes. Modeling/analyzing shape differences across subgroups in a population requires a consistent representation of all shapes in the entire population. Grooming (data preprocessing) and shape model optimization should be agnostic to those subgroups (i.e., fully unsupervised). It is the post analysis (after building the shape model) where the subgroups comes to play.

- I read that statistical analysis are usually done on PCA loading values using Hotelling T2 for group differences in another software (like Matlab). However, I don’t really understand the statistical analyses we should carry out… Could you please give us more information?  

You can use ShapeWorksView2 to visualize group differences. The release coming out next week will also include statistical group differences in Studio. You are welcomed to join our users meeting (remotely) next week to get more information about group differences computed via ShapeWorks.

The below paper also has more information and insights about hypothesis testing for shape models (in part implemented in ShapeWorks). 

- I wonder if a brief summary of ShapeWorks methodology exists. It could help me write the methodology part in our article. 
Yes. Please check out our GitHub readme where you can find a paragraph describing the underlying methodology. Also, I am attaching a book chapter that includes more details. Hope this helps.



thanks and best regards
Shireen


------------------------------------------
Shireen Elhabian, M.Sc., Ph.D.
Research Assistant Professor
School of Computing
Scientific Computing and Imaging Institute
University of Utah
WEB 2815
72 Central Campus Drive, Salt Lake City, UT, 84112
Phone (801) 587-3206
Fax (801) 585-6513
Home Page: http://www.sci.utah.edu/~shireen

Attachment: shapeworks_book_chapter_2017.pdf
Description: Adobe PDF document


On Jan 27, 2020, at 8:16 AM, quentin devignes <qdevignes@gmail.com> wrote:

Hi Shireen, 

First of all, let me wish you all the best for 2020! 

We managed to complete all the steps on our population using the python script. However, we have some questions about the next steps: 

- We carried out all the steps on our groups one after the other (we have 4 groups in our study). However, I wonder if a mean shape on all patients is needed for the following analyses (that means we should carry out all the steps on our entire population) 
- I read that statistical analysis are usually done on PCA loading values using Hotelling T2 for group differences in another software (like Matlab). However, I don’t really understand the statistical analyses we should carry out… Could you please give us more information?  
- I wonder if a brief summary of ShapeWorks methodology exists. It could help me write the methodology part in our article. 

Thanks and best regards, 

Quentin DEVIGNES
PhD student in neurosciences
Neuropsychologist
UMR-S 1172 Lille Neuroscience and Cognition - FRANCE

Le 14 déc. 2019 à 03:42, Shireen Elhabian <shireen@sci.utah.edu> a écrit :

Hi Quentin,

Nrrd files that are loaded in Studio should have the same voxel spacing and dimensions. Could you elaborate more on your grooming pipeline?

thanks and best regards
Shireen


------------------------------------------
Shireen Elhabian, M.Sc., Ph.D.
Research Assistant Professor
School of Computing
Scientific Computing and Imaging Institute
University of Utah
WEB 2815
72 Central Campus Drive, Salt Lake City, UT, 84112
Phone (801) 587-3206
Fax (801) 585-6513
Home Page: http://www.sci.utah.edu/~shireen

On Dec 10, 2019, at 9:53 AM, quentin devignes <qdevignes@gmail.com> wrote:

Hi Shireen, 

An error occurred when I tried to load post-grooming .nrrd images (see attached screenshot) for optimize step. I don’t understand why this error occurred while I have used exactly the same parameters for grooming… Could you please help me again? 

Thanks and best regards, 

Quentin DEVIGNES
PhD student in neurosciences
Neuropsychologist
INSERM U1171 - Lille, FRANCE


<Capture d’écran 2019-12-10 à 17.41.10.png>

Le 9 déc. 2019 à 14:20, quentin devignes <qdevignes@gmail.com> a écrit :

Hi Shireen, 

Thanks again for your help and advice. I wonder if ShapeWorks has already been used to study subcortical structures (caudate nucleus, hippocampus, etc.). Do you have any reference? 

Moreover, ShapeWorks does not provide any statistical analysis (between-group comparisons, for example). Which statistical software and/or method do you recommend? 

Best regards, 

Quentin DEVIGNES
PhD student in neurosciences
Neuropsychologist
INSERM U1171 - Lille, FRANCE

Le 8 nov. 2019 à 16:52, Shireen Elhabian <shireen@sci.utah.edu> a écrit :

Hi Quentin,

The current grooming steps in Studio don’t assume that you process all your data at once. You could groom a small subset at a time. However, to build a shape model (the Optimize stage), you would need to load all the groomed shapes in Studio. Another route would be considering the python scripts.

I would start with a small representative subset to tune groom/optimize/analyze parameters and then move forward with the larger dataset. Also, if you are willing to upload your data to the ShapeWorks portal, we could help you running the workflow on our servers.

thanks and best regards
Shireen


------------------------------------------
Shireen Elhabian, M.Sc., Ph.D.
Research Assistant Professor
School of Computing
Scientific Computing and Imaging Institute
University of Utah
WEB 2815
72 Central Campus Drive, Salt Lake City, UT, 84112
Phone (801) 587-3206
Fax (801) 585-6513
Home Page: http://www.sci.utah.edu/~shireen

On Nov 7, 2019, at 8:07 AM, quentin devignes <qdevignes@gmail.com> wrote:

Hi Shireen, 

I wonder if I have to groom every structures together or not. Indeed, I have 114 patients and my computer is not efficient enough to groom all patients together… Moreover, I wonder if this issue will pose a problem for the next steps… 

Thank you in advance for your answer, 

Best regards, 
Quentin DEVIGNES
PhD student in neurosciences
Neuropsychologist
INSERM U1171 - Lille, FRANCE

Le 6 nov. 2019 à 19:16, Shireen Elhabian <shireen@sci.utah.edu> a écrit :

Hi Quentin,

Thanks for trying out ShapeWorks for your analysis. 

What you are observing in the groomed data is mainly due to the blurring step. Since the caudate is a small structure and the voxel spacing is not small enough to have enough image support, the blur starts to erode (eat out) the underlying geometry. To mitigate this, you could trying one or more of the following:
- reduce the sigma. 
- resample the segmentations using a smaller voxel spacing (there is a command line tool for this in ShapeWorks tools, please checkout the ellipsoid example in ShapeWorks/Examples/Python/ellipsoidMain.py 
- use the “TopologyPreservingSmoothing” command line tool for smoothing rather than the blur tool currently in Studio.

You could also modify the ellipsoid example  to groom your data (which does resampling and the TopologyPreservingSmoothing).

Hope this helps.  

thanks and best regards
Shireen


------------------------------------------
Shireen Elhabian, M.Sc., Ph.D.
Research Assistant Professor
School of Computing
Scientific Computing and Imaging Institute
University of Utah
WEB 2815
72 Central Campus Drive, Salt Lake City, UT, 84112
Phone (801) 587-3206
Fax (801) 585-6513
Home Page: http://www.sci.utah.edu/~shireen

On Nov 5, 2019, at 10:50 AM, quentin devignes <qdevignes@gmail.com> wrote:

Dear Dr Elhabian, 

PhD student working on MRI features in Parkinson’s disease, I would like to use ShapeWorks to compare caudate nuclei and hippocampi shapes between four groups of patients. Previously, I had used SPHARM-PDM, but I read that ShapeWorks could be more relevant for such analyses. However, I have some issues with this software.

After having resampled images, I have loaded them in ShapeWorksStudio and launched the groom procedure. However, it failed if I put the « centre » parameter. So, I have relaunched it without this option and I have obtained groomed images. However, the new images seem to be quite different from the original images. I don’t know if I have to adjust options values for pad, antialias and blur (I let the default values). I did not find any relevant information online… Could you please help me with this issue? 

Some original images of left caudate nucleus and screenshots are attached in this email. 

Best regards, 

Quentin DEVIGNES
PhD student in neurosciences
Neuropsychologist
INSERM U1171 - Lille, FRANCE

<rc_u_G1_sub_100269SD100714_LeftCaudate.mha>
<rc_u_G1_sub_380304JM130613_LeftCaudate.mha>
<rc_u_G1_sub_430702JG100214_LeftCaudate.mha>
<Screenshot_GroomedImages.png>
<Screenshot_OriginalImages.png>











  • Re: [shapeworks-users] Questions about ShapeWorks, Shireen Elhabian, 01/31/2020

Archive powered by MHonArc 2.6.18.

Top of page