External Email - Use Caution
Dear experts,
We are working with a large MRI dataset (FLAIR, T1) and we are performing SAMSEG pipeline to estimate lesion volumes. We computed the registration between the two input images and the resampling as advised in the SAMSEG page (mri_coreg and mri_vol2vol) which worked perfectly. However, when we do the visual chek on the segmentation after running SAMSEG, it is completely wrong for some subjects. By looking at corresponding mode01_bias_corrected.mgz and mode02_bias_corrected.mgz images, they are too cropped (centered on the cerebellum) compared to the original ones (3D_T1 .nii.gz and 3D_FLAIR.nii.gz). We also noticed a < RuntimeWarning: invalid value encountered in true_divide > during the burn-in phase (in the GMM.py code).
Do you have any clue of what is happening with this cropping (can it be linked to the bias field estimation ?) and how we can correct it ?
Thanks for your help and I can provide some screenshots of the cropped images to help visualize the issue.
Best,
Achille Teillac
Not sure. The first step would be to check the affine registration with freeview template_coregistered.mgz input where input is the input to samseg
On 8/31/2021 5:27 AM, achille.teillac@chu-lyon.fr wrote:
External Email - Use Caution
Dear experts,
We are working with a large MRI dataset (FLAIR, T1) and we are performing SAMSEG pipeline to estimate lesion volumes. We computed the registration between the two input images and the resampling as advised in the SAMSEG page (/mri_coreg/ and /mri_vol2vol/) which worked perfectly. However, when we do the visual chek on the segmentation after running SAMSEG, it is completely wrong for some subjects. By looking at corresponding *mode01_bias_corrected.mgz *and *mode02_bias_corrected.mgz* images, they are too cropped (centered on the cerebellum) compared to the original ones (3D_T1 .nii.gz and 3D_FLAIR.nii.gz). We also noticed a « RuntimeWarning: invalid value encountered in true_divide » during the burn-in phase (in the GMM.py code).
Do you have any clue of what is happening with this cropping (can it be linked to the bias field estimation ?) and how we can correct it ?
Thanks for your help and I can provide some screenshots of the cropped images to help visualize the issue.
Best,
Achille Teillac
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
freesurfer@nmr.mgh.harvard.edu