Dear Kenji,
you can ignore the warnings, but have to keep in mind that the segmentation quality of your uncorrected data will be degraded compared to the corrected MP2RAGE data. The larger Euler Number might be caused by the strong local intensity inhomogeneities that are more difficult to correct.
In order to save the defects please try the attached version.
Best,
Christian
On Thu, 6 May 2021 23:21:10 +0900, Kenji Yoshimura <[log in to unmask]> wrote:
>Dear Christian,
>
>Thank you for your advice for using p0 labels.
>We re-ran segmentation pipeline with uncorrected MP2RAGE images masked by
>p0-labels which were obtained from denoised MP2RAGE segmentation, and we
>could get pretty good results!
>However, during segmentation, we got two warnings; (1)fine affine
>registration failed, and (2)SLC_noExpDef as shown below.
>Can these two warnings, especially fine affine registration failure, have
>some influence on the results?
>
>In addition, compared to the results from denoised MP2RAGE, surface
>topology defect and Euler number tend to be large. We suspect that
>insufficient masking in basal frontal and temporal caused poor surface
>reconstruction. Since we would like to check the distribution of
>topological defects, can we get maps that store where the topological
>defects are located? We set cat.extopts.verb = 3 in cat_defaults.m, but
>were unable to obtain the maps (in expert mode, the same output files
>compared to cat.extopts.verb = 2, and in usual mode, an error occurred and
>segmentation stopped).
>
>Thanks in advance.
>Best regards,
>
>------------------------------------------------------------------------
>CAT12.7 r1742: 1/5: ./MP2UNI/NIFTI_Brain_beta2/msk_cMP2UNI_HC003_V1.nii
>------------------------------------------------------------------------
>SANLM denoising (medium): 132s
>APP: Rough bias correction:
> Initialize 19s
> Estimate background 20s
> Initial correction 20s
> Refine background 12s
> Final correction 16s
> Final scaling 18s
> 120s
>Affine registration 20s
>Affine registration 11s
>SPM preprocessing 1 (estimate 1):
> First fine affine registration failed.
> Use affine registration from previous step.
> Final fine affine registration failed.
> Use fine affine registration from previous step. 120s
>SPM preprocessing 1 (estimate 2): 119s
>SPM preprocessing 2 (write):
> Write Segmentation 73s
> Update Segmentation 54s
> Update Skull-Stripping 221s
> Update probability maps 70s
> 418s
>Global intensity correction: 37s
>SANLM denoising after intensity normalization (medium): 211s
>Fast registration 106s
>Local adaptive segmentation (LASstr=0.50):
> Prepare maps 15s
> Prepare partitions 7s
> Prepare segments (LASmod = 1.00) 73s
> Estimate local tissue thresholds (WM) 130s
> Estimate local tissue thresholds (GM) 130s
> Estimate local tissue thresholds (CSF/BG) 27s
> Intensity transformation 397s
> SANLM denoising after LAS (medium) 245s
> 245s
>ROI segmentation (partitioning):
> Atlas -> subject space 16s
> Major structures 2s
> Ventricle detection 6s
> Blood vessel detection 6s
> WMH detection (WMHCstr=0.50 > WMHCstr'=0.17) 14s
> Manual stroke lesion detection 0s
> Closing of deep structures 1s
> Side alignment 2s
> Final corrections 17s
> 66s
>
>
>---------------------------------------------------------------------------------------------------
> WARNING 01: cat_main1639:SLC_noExpDef
> SLC is deactivated but there are 4.44 cm of voxels with
>zero value inside the brain!
>
>---------------------------------------------------------------------------------------------------
>Blood vessel correction (BVCstr=0.50): 5s
>Amap using initial SPM12 segmentations (MRF filter strength 0.06): 79s
> AMAP peaks: [CSF,GM,WM] = [0.41±0.06,0.68±0.09,0.98±0.04]
>Final cleanup (gcutstr=0.25):
> Level 1 cleanup (ROI estimation) 19s
> Level 1 cleanup (brain masking) 8s
> Level 2 cleanup (CSF correction) 5s
> Level 3 cleanup (CSF/WM PVE) 13s
> 45s
>Optimized Shooting registration with 2.50:-0.25:1.50 mm (regstr=0.50):
> Template:
>"/Users/yoshiken/spm12/toolbox/cat12/templates_volumes/Template_0_IXI555_MNI152_GS.nii"
> 1 | 2.50 | 0.0720 0.0000 0.0720
> 2 | 2.50 | 0.0703 0.0007 0.0710
> 3 | 2.50 | 0.0696 0.0011 0.0708
> 16 | 2.25 | 0.0737 0.0009 0.0745
> 17 | 2.25 | 0.0621 0.0033 0.0654
> 18 | 2.25 | 0.0592 0.0040 0.0632
> 30 | 2.00 | 0.0610 0.0055 0.0665
> 31 | 2.00 | 0.0555 0.0059 0.0614
> 32 | 2.00 | 0.0541 0.0066 0.0607
> 44 | 1.75 | 0.0543 0.0077 0.0620
> 45 | 1.75 | 0.0510 0.0078 0.0588
> 46 | 1.75 | 0.0502 0.0082 0.0584
> 58 | 1.50 | 0.0522 0.0088 0.0611
> 59 | 1.50 | 0.0475 0.0097 0.0572
> 60 | 1.50 | 0.0465 0.0103 0.0569
>Shooting registration with 2.50:-0.25:1.50 mm takes: 329s
> Prepare output 41s
> 371s
> Jacobian determinant (RMS): 0.007 0.072 0.114 0.130 0.149 | 0.153041
> Template Matching: 0.209 0.178 0.162 0.151 0.140 | 0.139592
>Write result maps: 45s
>Surface and thickness estimation:
>lh:
> Thickness estimation (0.50 mm):
> WM distance: 56s
> CSF distance: 28s
> PBT2x thickness: 17s
> 117s
> Create initial surface 284s
> Topology correction and surface refinement: 337s
> Correction of central surface in highly folded areas 12s
> Refine central surface 171s
> Correction of central surface in highly folded areas 2 19s
> Spherical mapping with areal smoothing 258s
> Spherical registration 469s
> Euler number / defect number / defect size: -50 / 17 / 1.39%
>
>rh:
> Thickness estimation (0.50 mm):
> WM distance: 60s
> CSF distance: 28s
> PBT2x thickness: 16s
> 119s
> Create initial surface 271s
> Topology correction and surface refinement: 334s
> Correction of central surface in highly folded areas 13s
> Refine central surface 170s
> Correction of central surface in highly folded areas 2 20s
> Spherical mapping with areal smoothing 348s
> Spherical registration 610s
> Euler number / defect number / defect size: -90 / 22 / 2.60%
>
>Final surface processing results:
> Average thickness: 2.5055 ± 0.8753 mm
> Surface intensity / position RMSE: 0.0730 / 0.1034
> Euler number / defectnumber / defect size: 74.0 / 19.5 / 2.00%
> Display thickness:
>/Users/yohsiken/PDcohort/B1corrected/Visit1/MP2UNI/NIFTI_Brain_beta2/surf/lh.thickness.msk_cMP2UNI_HC003_V1
> Display thickness:
>/Users/yohsiken/PDcohort/B1corrected/Visit1/MP2UNI/NIFTI_Brain_beta2/surf/rh.thickness.msk_cMP2UNI_HC003_V1
> Surface ROI estimation: Surface
>and thickness estimation takes: 3739s
>ROI estimation in native space:
> ROI estimation of 'cobra' atlas 44s
> ROI estimation of 'aal3' atlas 68s
> ROI estimation of 'anatomy3' atlas 147s
> ROI estimation of 'julichbrain' atlas 96s
> ROI estimation of 'BNAtlas_246' atlas 242s
> ROI estimation of 'neuromorphometrics' atlas 80s
> ROI estimation of 'lpba40' atlas 26s
> ROI estimation of 'hammers' atlas 75s
> ROI estimation of 'ibsr' atlas 21s
> ROI estimation of 'mori' atlas 69s
> Write results 72s
> 872s
>Quality check: 30s
>Developer display mode!
>Print 'Graphics' figure to:
>
>/Users/yohsiken/PDcohort/B1corrected/Visit1/MP2UNI/NIFTI_Brain_beta2/report/catreport_msk_cMP2UNI_HC003_V1.pdf
>
>------------------------------------------------------------------------
>CAT preprocessing takes 123 minute(s) and 14 second(s).
>Image Quality Rating (IQR): 90.37% (A-)
>Segmentations are saved in
>/Users/yohsiken/PDcohort/B1corrected/Visit1/MP2UNI/NIFTI_Brain_beta2/mri
>Reports are saved in
>/Users/yohsiken/PDcohort/B1corrected/Visit1/MP2UNI/NIFTI_Brain_beta2/report
>Labels are saved in
>/Users/yohsiken/PDcohort/B1corrected/Visit1/MP2UNI/NIFTI_Brain_beta2/label
>------------------------------------------------------------------------
>
>=============================================
>Department of Neurology
>Graduate School of Medicine, Kyoto University
>Kenji Yoshimura M.D.
>54 Shogoin-Kawahara-cho
>Sakyo-ku, Kyoto, Japan
>606-8507
>TEL: 075-751-3773
>FAX: 075-753-4257
>Mail: [log in to unmask]
>=============================================
>
>
>2021年4月29日(木) 5:05 Christian Gaser <[log in to unmask]>:
>
>> Dear Kenji,
>>
>> On 24 Apr 2021, at 0:29, Kenji Yoshimura wrote:
>>
>> > Dear Christian,
>> >
>> > Thank you very much for your suggestion! I didn't know Manual image
>> > masking
>> > tool, so I'll try it.
>> >
>> > Let me ask you one more question. We've run CAT12 segmentation with
>> > denoised-MP2RAGE, which gave fairly better results.
>> > If we use the output as masking, are binarized p0xxx.nii files the
>> > best for
>> > brain masks?
>> Yes, if you are interested in using the uncorrected MP2RAGE the masking
>> with the p0-labels will work.
>>
>> Best,
>>
>> Christian
>>
>> >
>> > Thanks in advance.
>> > Sincerely,
>> >
>> > =============================================
>> > Department of Neurology
>> > Graduate School of Medicine, Kyoto University
>> > Kenji Yoshimura M.D.
>> > 54 Shogoin-Kawahara-cho
>> > Sakyo-ku, Kyoto, Japan
>> > 606-8507
>> > TEL: 075-751-3773
>> > FAX: 075-753-4257
>> > Mail: [log in to unmask]
>> > =============================================
>> >
>> >
>> > 2021年4月23日(金) 16:43 Christian Gaser
>> > <[log in to unmask]>:
>> >
>> >> Dear Kenji,
>> >>
>> >> CAT12 is not yet very well prepared for that uncorrected 7T MP2RAGE
>> >> data
>> >> because the bias effects are usually too strong and in combination
>> >> with the
>> >> salt and pepper noise in the background that indeed confuses the
>> >> segmentation. You can try to mask your images with an external
>> >> obtained
>> >> mask (you can also try SPM12 segmentation) with this batch:
>> >> CAT12 > Tools > Manual image (lesion) masking)
>> >>
>> >> However, I know this is far from being perfect for that kind of data.
>> >> If
>> >> we have a few data for testing we can try to suggest a processing
>> >> strategy.
>> >>
>> >> Best,
>> >>
>> >> Christian
>> >>
>> >> On Thu, 22 Apr 2021 12:45:02 +0900, Kenji Yoshimura <
>> >> [log in to unmask]> wrote:
>> >>
>> >>> Dear Christian and CAT12 experts,
>> >>>
>> >>> We are trying to perform surface analysis with 7T-MP2RAGE.
>> >>> Original MP2RAGE (UNI-image) shows salt & pepper noise in the
>> >>> background
>> >>> and mastoid air cell/paranasal sinus due to its technological
>> >>> instability.
>> >>> Since this noise confuses tissue segmentation and skull stripping
>> >>> (especially in basal frontal and temporal lobes), the quality of
>> >>> surface
>> >>> reconstruction with UNI-image gets quite bad.
>> >>>
>> >>> We could obtain brain masks using external programs (e.g. FSL-BET),
>> >>> and
>> >>> this mask removes above noise.
>> >>> My question is, is it possible to use these own brain masks in CAT12
>> >>> segmentation pipeline? Or, is it possible to create brain masks
>> >>> using
>> >>> another T1WI-like image (MP2RAGE-Inv2) during UNI-image
>> >>> segmentation?
>> >>>
>> >>> Of course I can run segmentation with denoised-MP2RAGE (UNI-DEN;
>> >>> O’Brien
>> >>> KR, et al. PLoS One. 2014.), but our UNI-DEN images have a lot of
>> >>> bias,
>> >> and
>> >>> I want to check original (noise-rich) UNI-images...
>> >>>
>> >>> Thanks in advance.
>> >>> Best regards,
>> >>>
>> >>>
>> >>> =============================================
>> >>> Department of Neurology
>> >>> Graduate School of Medicine, Kyoto University
>> >>> Kenji Yoshimura M.D.
>> >>> 54 Shogoin-Kawahara-cho
>> >>> Sakyo-ku, Kyoto, Japan
>> >>> 606-8507
>> >>> TEL: 075-751-3773
>> >>> FAX: 075-753-4257
>> >>> Mail: [log in to unmask]
>> >>> =============================================
>> >>>
>> >>
>> >>
>> >>
>>
>>
>>
>
|