dke-questions > Tensor fit and registration
Showing 1-5 of 5 posts
Jan 26, 2018 01:01 AM | Meg Spriggs - University of Auckland
Tensor fit and registration
Hello,
I am currently trying to set up my first diffusion kurtosis imaging pipeline using DKE and DKE_FT (thank you to the creators for these awesome tools!) and I have two questions.
Firstly, looking at the output from DKE, there is a bit of noise left in my data. I have included denoising and Gibbs ringing correction in my preprocessing as well as FSLs eddy and top up. I am thinking about using the robust tensor fit option in DKE (i.e., RESTORE) to hopefully address remaining outliers, but am wondering what I should do with the other tensor fit options if I do this. i.e., should I continue using the default weighted constrained linear least squares method if I also include the robust option? Do you think that this is a good option for cleaning up the data, or should I include a different method of outlier detection somewhere else in my analysis?
Secondly, After putting my preprocessed images through DKE, I am hoping to do 2 things: run the FA/kurtosis images through TBSS, and run tractography. For TBSS, the images are registered to standard space within FSL's TBSS functions, which is great. But my question is whether I should also be registering the images to standard space before I run DKE_FT, and what tool you would recommend for doing this.
I apologise if these is a silly questions and appreciate any advice you may have.
Thank you for your help!
Kindest Regards,
Meg
I am currently trying to set up my first diffusion kurtosis imaging pipeline using DKE and DKE_FT (thank you to the creators for these awesome tools!) and I have two questions.
Firstly, looking at the output from DKE, there is a bit of noise left in my data. I have included denoising and Gibbs ringing correction in my preprocessing as well as FSLs eddy and top up. I am thinking about using the robust tensor fit option in DKE (i.e., RESTORE) to hopefully address remaining outliers, but am wondering what I should do with the other tensor fit options if I do this. i.e., should I continue using the default weighted constrained linear least squares method if I also include the robust option? Do you think that this is a good option for cleaning up the data, or should I include a different method of outlier detection somewhere else in my analysis?
Secondly, After putting my preprocessed images through DKE, I am hoping to do 2 things: run the FA/kurtosis images through TBSS, and run tractography. For TBSS, the images are registered to standard space within FSL's TBSS functions, which is great. But my question is whether I should also be registering the images to standard space before I run DKE_FT, and what tool you would recommend for doing this.
I apologise if these is a silly questions and appreciate any advice you may have.
Thank you for your help!
Kindest Regards,
Meg
Jan 30, 2018 04:01 PM | Corinne McGill - MUSC
RE: Tensor fit and registration
Hello Meg,
Thank you for using DKE!
I am not familiar with your data, but I normally would not recommend using the robust tensor fit option. Would you mind sending me your dke parameters text file?
In regard to your second question, you should not register images to standard space before running DKE and DKE_FT.
Thanks,
Corinne McGill
Thank you for using DKE!
I am not familiar with your data, but I normally would not recommend using the robust tensor fit option. Would you mind sending me your dke parameters text file?
In regard to your second question, you should not register images to standard space before running DKE and DKE_FT.
Thanks,
Corinne McGill
Jan 31, 2018 05:01 AM | Meg Spriggs - University of Auckland
RE: Tensor fit and registration
Hello Corinne,
Thank you for your reply.
Okay great. So I have separate parameter files for each participant so that the b-matrix is correctly rotated for tractography. But I have attached an example. I am currently using the weighted constrained linear least squares method for tensor fit.
Kindest regards,
Meg
Thank you for your reply.
Okay great. So I have separate parameter files for each participant so that the b-matrix is correctly rotated for tractography. But I have attached an example. I am currently using the weighted constrained linear least squares method for tensor fit.
Kindest regards,
Meg
Feb 1, 2018 02:02 PM | Emilie McKinnon - MUSC
RE: Tensor fit and registration
Hi Meg,
It looks like you have lots separate gradient direction files (with a low number of directions). Seeing that you use denoising and Gaussian smoothing and still have bad fits, I suspect that your gradient directions might be too dependent on one another for accurate fitting. The best way to test this is to look at the condition number of the matrix used for the weighted linear least squares ( if this matrix is ill-conditioned, things will go wrong). I understand that this is not a straightforward thing to do without the source code. However, we are working on that!
Just to make sure. Can you send an example image with "bad voxels" that you are trying to get rid of?
I hope this helps,
Emilie
It looks like you have lots separate gradient direction files (with a low number of directions). Seeing that you use denoising and Gaussian smoothing and still have bad fits, I suspect that your gradient directions might be too dependent on one another for accurate fitting. The best way to test this is to look at the condition number of the matrix used for the weighted linear least squares ( if this matrix is ill-conditioned, things will go wrong). I understand that this is not a straightforward thing to do without the source code. However, we are working on that!
Just to make sure. Can you send an example image with "bad voxels" that you are trying to get rid of?
I hope this helps,
Emilie
Feb 1, 2018 10:02 PM | Meg Spriggs - University of Auckland
RE: Tensor fit and registration
Hello Emilie,
Yes the b values vary around 0, 1000, and 2000 because they account for the imaging gradient. I thought it would be better to include this in the model, but would I be better to just use 0, 1000 and 2000?
Sorry for the potentially silly questions, but is there a way to test the model fit? Like a goodness-of-fit measure?
Thank you,
Meg
Yes the b values vary around 0, 1000, and 2000 because they account for the imaging gradient. I thought it would be better to include this in the model, but would I be better to just use 0, 1000 and 2000?
Sorry for the potentially silly questions, but is there a way to test the model fit? Like a goodness-of-fit measure?
Thank you,
Meg