FIX: Revise the reproducibility of CompCor masks #2130

oesteban · 2020-05-20T23:46:10Z

This PR revises the implementation of CompCor masks in an attempt to make it closer to the original proposal and at the same time address some recurring problems of fMRIPrep's tCompCor implementation.

Finally, with the more careful resampling of prior knowledge from the anatomical scan, this refactor should also make the aCompCor components more run-to-run repeatable.

aCompCor

The massaging of CompCor masks is now done in anatomical space where it is more precise, and a careful resampling to BOLD space follows.

The implementation deviates from Behzadi et al. Their original implementation thresholded the CSF and the WM partial-volume masks at 0.99 (i.e., 99% of the voxel volume is filled with a particular tissue), and then binary eroded that 2 voxels:

Anatomical data were segmented into gray matter, white matter, and CSF partial volume maps using the FAST algorithm available in the FSL software package (Smith et al., 2004). Tissue partial volume maps were linearly interpolated to the resolution of the functional data series using AFNI (Cox, 1996). In order to form white matter ROIs, the white matter partial volume maps were thresholded at a partial volume fraction of 0.99 and then eroded by two voxels in each direction to further minimize partial voluming with gray matter. CSF voxels were determined by first thresholding the CSF partial volume maps at 0.99 and then applying a threedimensional nearest neighbor criteria to minimize multiple tissue partial voluming. Since CSF regions are typically small compared to white matter regions mask, erosion was not applied.

This particular procedure is not generalizable to BOLD data with different voxel zooms as the mathematical morphology operations will be scaled by those. Also, from reading the excerpt above and the tCompCor description, I (@oesteban)
believe that they always operated slice-wise given the large slice-thickness of their functional data.

Instead, fMRIPrep's implementation deviates from Behzadi's implementation on two aspects:

the masks are prepared in high-resolution, anatomical space and then projected into BOLD space; and,
instead of using binary erosion, a dilated GM map is generated -- thresholding the corresponding PV map at 0.05 (i.e., pixels containing at least 5% of GM tissue) and then subtracting that map from the CSF, WM and CSF+WM (combined) masks. This should be equivalent to eroding the masks, except that the erosion only happens at direct interfaces with GM.

When the probseg maps provene from FreeSurfer's recon-all (i.e., they are discrete), binary maps are transformed into some sort of partial volume maps by means of a Gaussian smoothing filter with sigma adjusted by the size of the
BOLD data.

tCompCor

In the case of tCompCor, this commit removes the heavy erosion of the brain mask because 1) that wasn't part of the original proposal by Behzadi et al., and 2) the erosion was the potential source of errors from numpy complaining that it can't take from an empty axis of an array.

Based on these results, we chose a 2% threshold (∼20–30 voxels per slice) as a reasonable empirical threshold that effectively identified voxels with the highest fractional variance of physiological noise.

Although they do the calculation slice-wise, this commit rolls tCompCor back to calculate the 2% threshold on the whole-brain mask.

I would appreciate a deep review from @rciric & @jdkent to ensure my interpretation of Behzadi's paper is correct. I'm also tagging @effigies for a more top-level check of the changes. Further feedback from others is of course very much appreciated.

Resolves: #2129.
References: #2052.

pep8speaks · 2020-06-18T05:38:50Z

Hello @oesteban! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

There are currently no PEP 8 issues detected in this Pull Request. Cheers! 🍻

Comment last updated at 2020-08-28 08:00:36 UTC

oesteban · 2020-06-18T22:06:17Z

Ready for you @rciric

oesteban · 2020-06-20T15:43:02Z

The horrible contours on our subsampled dataset will be addressed by nipreps/niworkflows#544.

mgxd

👍 looks great

mgxd · 2020-06-22T13:24:21Z

fmriprep/utils/confounds.py

+    sigma = 0.5 * (zooms / imgzooms)
+    if len(sigma) > 1:
+        sigma = tuple(sigma)


is this check necessary? (will sigma ever be <= 1?)

rciric · 2020-07-08T12:31:34Z

Hello,

Apologies for the late response. I agree that morphological erosion using an isotropic fixed-voxel-size structuring element would generalize poorly to new resolutions. Reading the Behzadi paper again, I’m somewhat confused about the decision to warp the data into functional space before applying the erosion, and the motivation for doing so is unclear to me. I completely agree with the decision to perform the erosion in anatomical space, and the adaptive kernel size is a great idea.

Eroding using a low-threshold GM mask intuitively sounds like an improvement over conventional methods, but I do think that we should ideally examine the correlations of WM and CSF time series with whole-brain global signal when using the new approach to determine whether there are any major changes from the conventional approach. In particular, we would expect that limiting partial volume effects keeps the tissue-specific correlation with global signal reasonably low. This is of particular concern for those researchers who believe strongly that global mean signal should not be discarded (although I do not find myself in that camp).

My recommendation for tCompCor is similar: the proposed approach sounds very reasonable, although I think it wouldn't hurt to more clearly characterise the noise ROI and the component time series correlations at some point. tCompCor appears to be used with somewhat less frequency than aCompCor.

To verify, are the new aCompCor masks also going to define the regions over which the (global) mean WM and CSF signals are extracted?

The literature is not entirely uniform in use of anatomical CompCor approaches. In addition to the original method from Behzadi and colleagues, a similar aCompCor methodology proposed by Muschelli and colleagues is used with considerable frequency. In particular, the aCompCor evaluated by our group (and by Parkes et al. to my knowledge) more closely matched the Muschelli approach. By computing separate decompositions for each compartment mask and for the union mask, fMRIPrep already supports the most significant difference between the approaches. It's not, however, immediately clear to me whether Muschelli and colleagues performed their mask erosions in native anatomical space or in a standard space, since the maps they used were provided by a combined segmentation-normalisation procedure from SPM:

Subject-specific tissue probability maps generated during SPM’s unified segmentation-normalization process were restricted using a 99% probability threshold. To further reduce the risk of capturing signals of interest from adjacent grey matter voxels, the resulting WM mask was eroded using MATLAB’s imerode function, and the CSF mask was constrained to areas within the ALVIN mask of the ventricles (Kempton et al., 2011).

As a potential aid for evaluating the masks we obtain, I'm including here for reference the subject-wise voxel inclusion frequency map for WM and CSF masks in the Muschelli study.

(from Muschelli et al., 2014)

Also potentially of note, Power and colleagues performed their mask erosion using FreeSurfer. I'm not sure whether this erosion was performed in volumetric or surface space, and I only have a passing familiarity with FreeSurfer. In any case, there is substantial precedent for eroding masks in subject anatomical space. This approach was most similar to what our group did in the benchmarking paper, although we performed our erosions morphologically in volumetric space.

Each subject’s MP-RAGE underwent segmentation by FreeSurfer (version 5.0). Segments corresponding to the lateral ventricles, the white matter, the gray matter, and all within-brain voxels were extracted. 5 versions of the white matter and ventricular masks were formed: uneroded, and erosions 1–4 times. All masks were resliced to 3 mm isotropic dimensions to match the BOLD data. Maximal erosion (4×) masks were preferred; if maximal erosion yielded empty masks, lesser erosions were progressively considered until a mask with qualifying voxels was found. This relaxation occurred infrequently for white matter masks. Erosions of 1 were often used for CSF masks. These masks were used to extract nuisance signals, to calculate DV and SD within the masks, and to define the voxel timeseries displayed in Part I.

The description in a later work suggests to me that the erosion was likely also in volumetric space, although I again cannot say with confidence. This study also includes an example of eroded masks (corresponding to the deepest WM colour below, although I seem to recall that many of the example masks I saw from the WashU group were restricted to deeper voxels still.)

(from Power et al., 2017)

Overall, I agree with the premises of this revision, although I believe that further practical characterisation of fMRIPrep's CompCor procedure, potentially in the form of empirical correlations and voxel inclusion maps, would be welcome at some future stage.

effigies · 2020-07-08T13:01:22Z

I don't think the FreeSurfer method described could apply to the surface, as there is no representation of white matter or CSF on the surface. It sounds like they used the aseg segmentation.

oesteban · 2020-08-10T07:18:27Z

Hi @rciric, happy to do this:

we should ideally examine the correlations of WM and CSF time series with whole-brain global signal when using the new approach to determine whether there are any major changes from the conventional approach. In particular, we would expect that limiting partial volume effects keeps the tissue-specific correlation with global signal reasonably low.

If I understand correctly and to say that this PR improves the situation, we should calculate the correlation of global signal with the time-series extracted from the different CompCor masks. Then, if these correlations are not significantly larger than those before this PR, we can say these new masks at least don't make worse masks for compcor. Is that right?

effigies

This looks overall reasonable. I'm assuming the Gaussian blur parameters are just heuristic, but if there's a principled justification, we should probably list it explicitly in a comment.

fmriprep/utils/confounds.py

fmriprep/workflows/bold/confounds.py

effigies · 2020-08-18T19:25:47Z

fmriprep/utils/confounds.py

+    return out_file
+
+
+def acompcor_masks(in_files, is_aseg=False, zooms=None):


I don't really like sticking major pieces of machinery into a utils module and having the interfaces be thin wrappers on a hidden function. I think at one point we were doing something like:

class AlgorithmInterface: def _run_interface(self, runtime): self._results = _algorithm(**self.inputs.get()) return runtime def _algorithm(arg1, arg2): ... return {"out1": ...}

And the justification was that it would be easier to test a function than an interface. We're not testing this, and now it's being hidden away into an unassuming corner of the code.

I don't want to hold up the PR, but I would suggest we think through our organization and apply it consistently.

I don't want to hold up the PR, but I would suggest we think through our organization and apply it consistently.

Yup, this would be good. Let's make sure we find a time after the LTS. At first my only comment would be that sphinx does not take private functions by default (that's why this is public). That said, if when we meet we decide to have everything together, fine by me - this was kind of an arbitrary decision.

oesteban

Thanks for the review :)

oesteban · 2020-08-18T19:39:41Z

fmriprep/utils/confounds.py

+    return out_file
+
+
+def acompcor_masks(in_files, is_aseg=False, zooms=None):


I don't want to hold up the PR, but I would suggest we think through our organization and apply it consistently.

Yup, this would be good. Let's make sure we find a time after the LTS. At first my only comment would be that sphinx does not take private functions by default (that's why this is public). That said, if when we meet we decide to have everything together, fine by me - this was kind of an arbitrary decision.

fmriprep/workflows/bold/confounds.py

effigies · 2020-08-18T21:16:42Z

The masks in reports look pretty bad.

effigies · 2020-08-19T15:30:17Z

Pushed your previous commit to upstream/compcormasks to generate reports. The good news is: my changes didn't cause the bad-looking reports.

oesteban · 2020-08-19T15:44:10Z

The masks in reports look pretty bad.

You mean for ds054 and ds210 (i.e., --fs-no-reconall), right?

I have mostly tested with recon-all - that's why I haven't lent much attention to ds054 and ds210. That said, I'm surprised that the combined masks look so bad.

To recapitulate:

Before this PR we were probably representing the combined mask (as stated in the caption of the reportlet), and it never ever contained CSF. I think our CSF mask was so exclusive that very few voxels were in it (and hence, the empty take errors from numpy). - I say this just in case you think ds005's reports look awful too.

Now, the masks with recon-all look pretty good. I've checked all of them after running ds114, and @Shotgunosine has also generated a good bunch (which looked good to me and @rciric will soon assess after his quals).

Then, we have the --fs-no-reconall path. I know I have run ds054 without reconall at full resolution and that gave me sound results. This might be a result of the very big pixels of the WM mask with the heavily subsampled dataset. Let me run a bunch of dataset. Perhaps, I should start with ds114 (and --fs-no-reconall), and see how different the masks are w.r.t. the default.

effigies · 2020-08-19T16:09:12Z

You're right, I didn't look at ds005 because it takes so much longer to complete.

Do the voxels near the edge of the brain mask occur in higher resolution images? Are they problematic? I assume they're classified as CSF, but I would worry you get GM signal bleed-over (possibly resulting in the high variance for tCompCor).

oesteban · 2020-08-19T16:16:50Z

Do the voxels near the edge of the brain mask occur in higher resolution images? Are they problematic? I assume they're classified as CSF, but I would worry you get GM signal bleed-over (possibly resulting in the high variance for tCompCor).

Are you referring to pink (aCompCor) or blue (tCompCor)? I guess they are not pink because the new implementation is quite careful - if they are selected, the certainty of them being CSF is very high.

Regarding tCompCor, this is expected (which does not mean good). I particularly asked @rciric about this, and he didn't seem too concerned. Please note that the original Behzadi paper DID NOT erode the brain mask (as we were doing), and also, their top percentile was calculated slice-wise (with veeery thick slices). My impression is that the original implementation of tCompCor wasn't great because of this.

We may want to come back to the previous implementation, but I definitely see the new tCompCor masks picking up voxels in the brain stem and external parts of the commissures - which is a good thing, I believe.

effigies · 2020-08-19T16:23:35Z

Are you referring to pink (aCompCor) or blue (tCompCor)?

They show up in both.

Fair enough.

oesteban · 2020-08-27T16:50:54Z

Okay, running ds054 and ds210 at their original resolution with --fs-no-reconall. Will report back when done.

@oesteban

This commit revises the implementation of CompCor masks in an attempt to make it closer to the original proposal and at the same time address some recurring problems of *fMRIPrep*'s tCompCor implementation. Finally, with the more careful resampling of prior knowledge from the anatomical scan, this refactor should also make the aCompCor components more run-to-run repeatable. aCompCor -------- The massaging of CompCor masks is now done in anatomical space where it is more precise, and a careful resampling to BOLD space follows. The implementation deviates from Behzadi et al. Their original implementation thresholded the CSF and the WM partial-volume masks at 0.99 (i.e., 99% of the voxel volume is filled with a particular tissue), and then binary eroded that 2 voxels: > Anatomical data were segmented into gray matter, white matter, > and CSF partial volume maps using the FAST algorithm available > in the FSL software package (Smith et al., 2004). Tissue partial > volume maps were linearly interpolated to the resolution of the > functional data series using AFNI (Cox, 1996). In order to form > white matter ROIs, the white matter partial volume maps were > thresholded at a partial volume fraction of 0.99 and then eroded by > two voxels in each direction to further minimize partial voluming > with gray matter. CSF voxels were determined by first thresholding > the CSF partial volume maps at 0.99 and then applying a threedimensional > nearest neighbor criteria to minimize multiple tissue > partial voluming. Since CSF regions are typically small compared > to white matter regions mask, erosion was not applied. This particular procedure is not generalizable to BOLD data with different voxel zooms as the mathematical morphology operations will be scaled by those. Also, from reading the excerpt above and the tCompCor description, I (@oesteban) believe that they always operated slice-wise given the large slice-thickness of their functional data. Instead, *fMRIPrep*'s implementation deviates from Behzadi's implementation on two aspects: * the masks are prepared in high-resolution, anatomical space and then projected into BOLD space; and, * instead of using binary erosion, a dilated GM map is generated -- thresholding the corresponding PV map at 0.05 (i.e., pixels containing at least 5% of GM tissue) and then subtracting that map from the CSF, WM and CSF+WM (combined) masks. This should be equivalent to eroding the masks, except that the erosion only happens at direct interfaces with GM. When the probseg maps provene from FreeSurfer's ``recon-all`` (i.e., they are discrete), binary maps are *transformed* into some sort of partial volume maps by means of a Gaussian smoothing filter with sigma adjusted by the size of the BOLD data. tCompCor -------- In the case of *tCompCor*, this commit removes the heavy erosion of the brain mask because 1) that wasn't part of the original proposal by Behzadi et al., and 2) the erosion was the potential source of errors from numpy complaining that it can't take from an empty axis of an array. > Based on these results, we chose a 2% threshold > (∼20–30 voxels per slice) as a reasonable empirical > threshold that effectively identified voxels with > the highest fractional variance of physiological noise. Although they do the calculation slice-wise, this commit rolls tCompCor back to calculate the 2% threshold on the whole brain mask. Resolves: nipreps#2129. References: nipreps#2052.

Co-authored-by: Chris Markiewicz <[email protected]>

oesteban · 2020-08-28T08:46:40Z

Okay, running ds054 and ds210 at their original resolution with --fs-no-reconall. Will report back when done.

Masks look okay (better with FS, but that's the drill). I'm building a new singularity image (which required cutting an RC of sMRIPrep) to test these masks including the fix from Chris to make the reportlet show the combined mask instead of the wm mask.

As soon as I have both datasets run with the latest image I'll ask @rciric for his trained eye.

Meanwhile, please have a look into #2245 - I think it would be worth pushing into 20.1.x to be able to do the correlation w/global-signal test.

effigies · 2020-09-02T12:57:20Z

@oesteban @rciric What's the status of this? I think we're very close to a release, so let me know if I can help on anything.

oesteban · 2020-09-02T14:38:13Z

My plan is to quickly make a PR to add the CompCor time series to the bar plot of correlations with the GS (and remove the components from there). That if @rciric thinks it's a good idea. Worst case scenario, the plots will become more readable.

…

On Wed, Sep 2, 2020, 14:57 Chris Markiewicz ***@***.***> wrote: @oesteban <https://github.com/oesteban> @rciric <https://github.com/rciric> What's the status of this? I think we're very close to a release, so let me know if I can help on anything. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#2130 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAESDRSHLSWESN4TUY4AXRDSDY6L7ANCNFSM4NGMUUGQ> .

effigies · 2020-09-02T15:05:58Z

Is this PR ready to merge, then?

oesteban · 2020-09-02T15:12:50Z

Yes, I think that's safe

…

On Wed, Sep 2, 2020, 17:06 Chris Markiewicz ***@***.***> wrote: Is this PR ready to merge, then? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#2130 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAESDRXJYUT22W32H2R5QNTSDZNOPANCNFSM4NGMUUGQ> .

20.2.0 RC0 (September 04, 2020) With this third minor release series of 2020, the first *fMRIPrep LTS* (*long-term support*) is finally here! This release contains a number of bug-fixes and enhancements mostly related to easing the maintenance, anticipating patch-release breaking changes to ensure a longstanding LTS, and addressing some run-to-run repeatability problems of the CompCor implementation. Thank you for using *fMRIPrep*! If you encounter any issues with this release, please let us know by posting an issue on our GitHub page! A full list of changes can be found below. * FIX: Revise the reproducibility of CompCor masks (nipreps#2130) * FIX: Simplify transform aggregation in resampling, pass identity transforms for multi-echo cases (nipreps#2239) * FIX: Skip the T1w check if ``--anat-derivatives`` is provided. (nipreps#2201) * FIX: Storing ``--bids-filters`` within config file (nipreps#2177) * FIX: Revise multi-echo reference generation, permitting using SBRefs too (nipreps#1803) * FIX: FreeSurfer license manipulation & canary * ENH: Output CompCor masks if ``--debug compcor`` is passed (nipreps#2248) * ENH: Conform to BIDS Derivatives as of BIDS 1.4.0 (nipreps#2223) * ENH: Reuse config (nipreps#2240) * ENH: Save BOLD-anatomical transforms to derivatives folder (nipreps#2233) * ENH: Leverage BIDSLayout's ``database_path`` (nipreps#2203) * ENH: Add ``--no-tty`` option to ``fmriprep-docker.py`` (nipreps#2204) * ENH: Report number of echoes in BOLD summary. (nipreps#2184) * ENH: Ensure NiPype telemetry is just pinged once (nipreps#2168) * DOC: Update reference in "Refinement of Brain Mask" description (nipreps#2215) * DOC: List *TemplateFlow* templates that need to be prefetched (nipreps#2196) * DOC: Update references to https://github.com/nipreps (nipreps#2191) * DOC: Pin NiPype with new Sphinx extension syntax (nipreps#2092) * MAINT: Track nipreps#2252 from 20.1.x series (nipreps#2253) * MAINT: Silence PyBIDS warning by setting extension mode (nipreps#2250) * MAINT: Drop CircleCI docs build (nipreps#2247) * MAINT: Pin latest *NiPreps* (nipreps#2244) * MAINT: Update ``setup.cfg`` (flake8 and pytest) (nipreps#2183) * MAINT: Delete release-drafter (nipreps#2169) * MAINT: Track bug-fix release on the 20.1.x series (nipreps#2165) * MAINT: Remove auto-comment bot (nipreps#2166) * MAINT: Improve the questions on the bug-report template (nipreps#2158)

20.2.0rc1 With this third minor release series of 2020, the first *fMRIPrep LTS* (*long-term support*) is finally here! This release contains a number of bug-fixes and enhancements mostly related to easing the maintenance, anticipating patch-release breaking changes to ensure a longstanding LTS, and addressing some run-to-run repeatability problems of the CompCor implementation. .. admonition:: Long-Term Support (LTS) *fMRIPrep* 20.2 LTS introduces the `long-term support program <https://www.nipreps.org/devs/releases/#long-term-support-series>`__. This LTS version will be kindly steered and maintained by the group of Dr. Basile Pinsard and Prof. Pierre Bellec at `CRIUGM <http://www.criugm.qc.ca/>`__, (Université de Montréal). The LTS is planned for a window of 4 years of support (i.e., until September 2024). .. caution:: As with all minor version increments, working directories from previous versions **should not be reused**. Thank you for using *fMRIPrep*! If you encounter any issues with this release, please let us know by posting an issue on our GitHub page! A full list of changes can be found below. * FIX: Get missing ``probseg`` file from MNI152NLin2009cAsym (nipreps#2271) * FIX: Restore ``--ignore t2w/flair`` options (nipreps#2260) * FIX: Revise the reproducibility of *CompCor* masks (nipreps#2130) * FIX: Simplify transform aggregation in resampling, pass identity transforms for multi-echo cases (nipreps#2239) * FIX: Skip the T1w check if ``--anat-derivatives`` is provided. (nipreps#2201) * FIX: Storing ``--bids-filters`` within config file (nipreps#2177) * FIX: Revise multi-echo reference generation, permitting using SBRefs too (nipreps#1803) * FIX: *FreeSurfer* license manipulation & canary * ENH: Output CompCor masks if ``--debug compcor`` is passed (nipreps#2248) * ENH: Conform to BIDS Derivatives as of BIDS 1.4.0 (nipreps#2223) * ENH: Reuse config (nipreps#2240) * ENH: Save BOLD-anatomical transforms to derivatives folder (nipreps#2233) * ENH: Leverage BIDSLayout's ``database_path`` (nipreps#2203) * ENH: Add ``--no-tty`` option to ``fmriprep-docker.py`` (nipreps#2204) * ENH: Report number of echoes in BOLD summary. (nipreps#2184) * ENH: Ensure *NiPype* telemetry is just pinged once (nipreps#2168) * DOC: Add FAQ entry for using pre-indexed layouts (nipreps#2256) * DOC: Update reference in "Refinement of Brain Mask" description (nipreps#2215) * DOC: List *TemplateFlow* templates that need to be prefetched (nipreps#2196) * DOC: Update references to https://github.com/nipreps (nipreps#2191) * DOC: Pin *NiPype* with new Sphinx extension syntax (nipreps#2092) * MAINT: Track nipreps#2269 and nipreps#2269, bug-fixes on the 20.1.x series * MAINT: Remove derivatives from layout index ignores (nipreps#2258) * MAINT: Track nipreps#2252 from 20.1.x series (nipreps#2253) * MAINT: Silence *PyBIDS* warning by setting extension mode (nipreps#2250) * MAINT: Drop CircleCI docs build (nipreps#2247) * MAINT: Pin latest *NiPreps* (nipreps#2244) * MAINT: Update ``setup.cfg`` (flake8 and pytest) (nipreps#2183) * MAINT: Delete release-drafter (nipreps#2169) * MAINT: Track bug-fix release on the 20.1.x series (nipreps#2165) * MAINT: Remove auto-comment bot (nipreps#2166) * MAINT: Improve the questions on the bug-report template (nipreps#2158) .. admonition:: Author list for papers based on *fMRIPrep* 20.2 LTS series As described in the `Contributor Guidelines <https://www.nipreps.org/community/CONTRIBUTING/#recognizing-contributions>`__, anyone listed as developer or contributor may write and submit manuscripts about *fMRIPrep*. To do so, please move the author(s) name(s) to the front of the following list: Markiewicz, Christopher J. \ :sup:`1`\ ; Goncalves, Mathias \ :sup:`1`\ ; DuPre, Elizabeth \ :sup:`2`\ ; Kent, James D. \ :sup:`3`\ ; Salo, Taylor \ :sup:`4`\ ; Ciric, Rastko \ :sup:`1`\ ; Pinsard, Basile \ :sup:`5`\ ; Finc, Karolina \ :sup:`6`\ ; de la Vega, Alejandro \ :sup:`7`\ ; Feingold, Franklin \ :sup:`1`\ ; Tooley, Ursula A. \ :sup:`8`\ ; Benson, Noah C. \ :sup:`9`\ ; Urchs, Sebastian \ :sup:`2`\ ; Blair, Ross W. \ :sup:`1`\ ; Erramuzpe, Asier \ :sup:`10`\ ; Lurie, Daniel J. \ :sup:`11`\ ; Heinsfeld, Anibal S. \ :sup:`12`\ ; Jacoby, Nir \ :sup:`13`\ ; Jamison, Keith W. \ :sup:`14`\ ; Frederick, Blaise B. \ :sup:`15, 16`\ ; Valabregue, Romain \ :sup:`17`\ ; Sneve, Markus H. \ :sup:`18`\ ; Liem, Franz \ :sup:`19`\ ; Adebimpe, Azeez \ :sup:`20`\ ; Velasco, Pablo \ :sup:`21`\ ; Wexler, Joseph B. \ :sup:`1`\ ; Groen, Iris I. A. \ :sup:`22`\ ; Ma, Feilong \ :sup:`23`\ ; Amlien, Inge K. \ :sup:`18`\ ; Bellec, Pierre \ :sup:`5`\ ; Cieslak, Matthew \ :sup:`20`\ ; Devenyi, Grabriel A. \ :sup:`24`\ ; Ghosh, Satrajit S. \ :sup:`25, 26`\ ; Gomez, Daniel E. P. \ :sup:`27`\ ; Halchenko, Yaroslav O. \ :sup:`23`\ ; Isik, Ayse Ilkay \ :sup:`28`\ ; Moodie, Craig A. \ :sup:`1`\ ; Naveau, Mikaël \ :sup:`29`\ ; Rivera-Dompenciel, Adriana \ :sup:`3`\ ; Satterthwaite, Theodore D. \ :sup:`20`\ ; Sitek, Kevin R. \ :sup:`30`\ ; Stojić, Hrvoje \ :sup:`31`\ ; Thompson, William H. \ :sup:`1`\ ; Wright, Jessey \ :sup:`1`\ ; Ye, Zhifang \ :sup:`32`\ ; Gorgolewski, Krzysztof J. \ :sup:`1`\ ; Poldrack, Russell A. \ :sup:`1`\ ; Esteban, Oscar \ :sup:`33`\ . Affiliations: 1. Department of Psychology, Stanford University 2. Montreal Neurological Institute, McGill University 3. Neuroscience Program, University of Iowa 4. Department of Psychology, Florida International University 5. SIMEXP Lab, CRIUGM, University of Montréal, Montréal, Canada 6. Centre for Modern Interdisciplinary Technologies, Nicolaus Copernicus University in Toruń 7. University of Texas at Austin 8. Department of Neuroscience, University of Pennsylvania, PA, USA 9. Department of Psychology, New York University 10. Computational Neuroimaging Lab, BioCruces Health Research Institute 11. Department of Psychology, University of California, Berkeley 12. Child Mind Institute 13. Department of Psychology, Columbia University 14. Department of Radiology, Weill Cornell Medicine 15. McLean Hospital Brain Imaging Center, MA, USA 16. Consolidated Department of Psychiatry, Harvard Medical School, MA, USA 17. CENIR, INSERM U1127, CNRS UMR 7225, UPMC Univ Paris 06 UMR S 1127, Institut du Cerveau et de la Moelle épinière, ICM, F-75013, Paris, France 18. Center for Lifespan Changes in Brain and Cognition, University of Oslo 19. URPP Dynamics of Healthy Aging, University of Zurich 20. Perelman School of Medicine, University of Pennsylvania, PA, USA 21. Center for Brain Imaging, New York University 22. Department of Psychology, New York University, NY, USA 23. Dartmouth College: Hanover, NH, United States 24. Department of Psychiatry, McGill University 25. McGovern Institute for Brain Research, MIT, MA, USA 26. Department of Otolaryngology, Harvard Medical School, MA, USA 27. Donders Institute for Brain, Cognition and Behaviour, Radboud University Nijmegen 28. Max Planck Institute for Empirical Aesthetics 29. Cyceron, UMS 3408 (CNRS - UCBN), France 30. Speech & Hearing Bioscience & Technology Program, Harvard University 31. Max Planck UCL Centre for Computational Psychiatry and Ageing Research, University College London 32. State Key Laboratory of Cognitive Neuroscience and Learning, Beijing Normal University 33. Department of Radiology, CHUV, Université de Lausanne

oesteban added the bug label May 20, 2020

This comment has been minimized.

Sign in to view

oesteban marked this pull request as draft May 20, 2020 23:46

oesteban force-pushed the fix/2129-compcor-masks-reliability branch 2 times, most recently from 8609326 to 692b15a Compare May 21, 2020 00:12

oesteban added this to the 20.2.0 LTS milestone May 22, 2020

oesteban force-pushed the fix/2129-compcor-masks-reliability branch from 692b15a to 37105c6 Compare May 28, 2020 23:25

oesteban marked this pull request as ready for review May 28, 2020 23:26

oesteban force-pushed the fix/2129-compcor-masks-reliability branch 3 times, most recently from 309640f to 39d0498 Compare June 12, 2020 22:50

oesteban mentioned this pull request Jun 14, 2020

FIX: Progress partial volume maps instead of posteriors (FSL FAST) nipreps/smriprep#213

Merged

oesteban force-pushed the fix/2129-compcor-masks-reliability branch from 39d0498 to e0baea4 Compare June 18, 2020 05:38

oesteban requested review from effigies, jdkent and rciric June 18, 2020 05:49

oesteban force-pushed the fix/2129-compcor-masks-reliability branch 3 times, most recently from 6bc0480 to 748d17e Compare June 18, 2020 19:09

oesteban mentioned this pull request Jun 19, 2020

Artifacts in mosaic plots when intersection of masks and visualization planes is minimal nipreps/niworkflows#542

Closed

mgxd reviewed Jun 22, 2020

View reviewed changes

mgxd mentioned this pull request Jul 17, 2020

CSF n/a in confounds file #2209

Closed

oesteban closed this Aug 10, 2020

oesteban reopened this Aug 10, 2020

effigies reviewed Aug 18, 2020

View reviewed changes

oesteban commented Aug 18, 2020

View reviewed changes

oesteban mentioned this pull request Aug 28, 2020

Write out "sanity" derivatives / introspection mode #2245

Closed

oesteban and others added 3 commits August 28, 2020 01:00

docs: update caption of BOLD ROIs reportlet and boilerplate

da14b63

Apply suggestions from code review

bafb714

Co-authored-by: Chris Markiewicz <[email protected]>

oesteban force-pushed the fix/2129-compcor-masks-reliability branch from 309264b to bafb714 Compare August 28, 2020 08:00

effigies mentioned this pull request Sep 1, 2020

ENH: Output compcor masks if --debug compcor is passed #2248

Merged

effigies merged commit da1e81f into nipreps:master Sep 2, 2020

oesteban deleted the fix/2129-compcor-masks-reliability branch September 3, 2020 15:54

		return out_file


		def acompcor_masks(in_files, is_aseg=False, zooms=None):

FIX: Revise the reproducibility of CompCor masks #2130

FIX: Revise the reproducibility of CompCor masks #2130

Uh oh!

Conversation

oesteban commented May 20, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

aCompCor

tCompCor

Uh oh!

This comment has been minimized.

pep8speaks commented Jun 18, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Comment last updated at 2020-08-28 08:00:36 UTC

Uh oh!

oesteban commented Jun 18, 2020

Uh oh!

oesteban commented Jun 20, 2020

Uh oh!

mgxd left a comment

Choose a reason for hiding this comment

Uh oh!

mgxd Jun 22, 2020

Choose a reason for hiding this comment

Uh oh!

rciric commented Jul 8, 2020

Uh oh!

effigies commented Jul 8, 2020

Uh oh!

oesteban commented Aug 10, 2020

Uh oh!

effigies left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

effigies Aug 18, 2020

Choose a reason for hiding this comment

Uh oh!

oesteban Aug 18, 2020

Choose a reason for hiding this comment

Uh oh!

oesteban left a comment

Choose a reason for hiding this comment

Uh oh!

oesteban Aug 18, 2020

Choose a reason for hiding this comment

Uh oh!

Uh oh!

effigies commented Aug 18, 2020

Uh oh!

effigies commented Aug 19, 2020

Uh oh!

oesteban commented Aug 19, 2020

Uh oh!

effigies commented Aug 19, 2020

Uh oh!

oesteban commented Aug 19, 2020

Uh oh!

effigies commented Aug 19, 2020

Uh oh!

oesteban commented Aug 27, 2020

Uh oh!

oesteban commented Aug 28, 2020

Uh oh!

effigies commented Sep 2, 2020

Uh oh!

oesteban commented Sep 2, 2020 via email

Uh oh!

effigies commented Sep 2, 2020

Uh oh!

oesteban commented Sep 2, 2020 via email

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

oesteban commented May 20, 2020 •

edited

Loading

pep8speaks commented Jun 18, 2020 •

edited

Loading