[9T8E] Pipeline reproduction (SPM deriv)

[Arshitha]

Softwares

SPM12 (7219)

Input data

derivatives (fMRIprep)

Additional context

Neurovault collection link: https://neurovault.org/collections/4870/
Pre-registration link: https://osf.io/54w2n
Softwares used: SPM12 (7219)
Number of participants: 106
Exclusion: sub-003 and sub-107

see description below for further details

List of tasks

Please tick the boxes below once the corresponding task is finished. 👍

• 👌 A maintainer of the project approved the issue, by assigning a 🏁status: ready for dev label to it.
• 🌳 Create a branch on your fork to start the reproduction.
• 🌅 Create a file team_{team_id}.py inside the narps_open/pipelines/ directory. You can use a file inside narps_open/pipelines/templates file as a template if needed.
• 🧠 Write the code for the pipeline, using Nipype and the file architecture described in docs/pipelines.md.
• 📘 Make sure your code is documented enough.
• 🐍 Make sure your code is explicit and conforms with PEP8.
• 🔬 Create tests for your pipeline. You can use files in tests/pipelines/test_team_* as examples.
• 🔬 Make sure your code passes all the tests you created (see docs/testing.md).
• 📥 create a pull request from your code.

---

NARPS team description : 9T8E

General

• teamID : 9T8E
• NV_collection_link : https://neurovault.org/collections/4870/
• results_comments : The maps added to neurovault were created using nibabel (i.e. combining .hdr and .img files), in this step NaNs were also removed from the images and replaced with 0s. Degrees of freedom and cluster level correction can be found in the descrption fields of the maps.
• preregistered : Yes
• link_preregistration_form : https://osf.io/54w2n
• regions_definition : Using restrictions on the brain regions by predefined ROIs created using neurosynths and calculating the overlap with the FWE corrected activation maps. The decision, however, could be refined by visual inspection.
• softwares : SPM12 (7219)
• general_comments : No voxels in the thresholded maps (FWE corrected) for hypotheses 5, 6, and 9 survived. To submit thresholded maps anyways, we submitted images consisting of 0s in the same dimensions of the unthresholded t-maps (also using the same header infos etc.). The images were created in nibabel. Information that no voxel survived can be found in the "Description" field on neurovault.

Exclusions

• n_participants : 106
• exclusions_details : sub-003, design matrix did include non-unique columns and could not be used for t-tests in SPM.
  sub-107, design matrix did include non-unique columns and could not be used for t-test in SPM

Preprocessing

• used_fmriprep_data : Yes
• preprocessing_order : After the fmriprep v1.1.4 processing (the provided data), we additionally applied spatial smoothing.
• brain_extraction : see fmriprep 1.1.4
• segmentation : see fmriprep 1.1.4
• slice_time_correction : non was perfomed (according to fmriprep 1.1.4 reports).
• motion_correction : see fmriprep 1.1.4
• motion :
• gradient_distortion_correction : see fmriprep 1.1.4
• intra_subject_coreg : see fmriprep 1.1.4
• distortion_correction : see fmriprep 1.1.4
• inter_subject_reg : see fmriprep 1.1.4
• intensity_correction : see fmriprep 1.1.4
• intensity_normalization : see fmriprep 1.1.4
• noise_removal : see fmriprep 1.1.4
• volume_censoring : see fmriprep 1.1.4
• spatial_smoothing : SPM12 (7219), 8mm FWHM kernel was applied to the images in MNI-space (the output of the functional pipeline of fmriprep v1.1.4).
• preprocessing_comments : Images had to be unzipped (Matlab9.1 gunzip) before they could be further processed in SPM12. After first-level estimation the smoothed and unzipped images were deleted to conserve disk-space.

Analysis

• data_submitted_to_model : We included 106 subjects in the final analysis.
  On a subject level all four runs werde modeled in a single design matrix. Each of the 4 runs consisted of 453 images. Meaning that 1812 timepoints were included in the analysis.
• spatial_region_modeled : Full-brain
• independent_vars_first_level : SPM12 (7219, if not other specified, defaults were used).

1. Accepted trials (pooling weakly and strongly accept), with a duration set to 4s:
   Parametric modulators:
   • reaction time
   • expected gain
   • expected loss
2. Rejected trials (pooling weakly and strongly reject), with a duration set to 4s:
   Parametric modulators:
   • reaction time
   • expected gain
   • expected loss
3. Trials with no responses (if present), duration of 4s.

We used a canonical HRF plus temporal derivatives in SPM12.
Head movement was accounted for by using the six movement regressors (translations and rotations), and framwise displacement as a non-linear combination of the movement parameters.
The default high-pass filter with a cutoff of 128s was applied.
Regressors were replicated for each run (or 'session' in SPM terms). I.e. regressors and their parametric modulations were estimated for each session.

• RT_modeling : pm
• movement_modeling : 1
• independent_vars_higher_level : For hypothesis 1 to 8 first level contrasts were submitted to a one-sample permutation t-test (SnPM). For hypothesis 9 a two-sample permutation t-tests was used to compare equal range - equal indifference groups.
  No other covariates were used.
• model_type : Mass univariate.
• model_settings : Random effects in SPM with AR(1) as drift model. Everything else following SPM12 defaults.
• inference_contrast_effect : For each subject we calculated two contrast: Parametrics effect of gain for accepted (Gain[Accept]) and rejected gambles (Gain[Reject]) and parametric effects of loss for accepted and rejected gambles. Each contrast consisted of 8 beta estimates. In case of parametric effects of gain the contrast consisted of (Gain[Accept] + Gain[Reject])_run1 + (Gain[Accept] + Gain[Reject] Gain)_run2 + (Gain[Accept] + Gain[Reject])_run3 + (Gain[Accept] + Gain[Reject])_run4. Contrasts were created using a simple replication ('repl') in the SPM contrast manager. No session specific scaling was applied.
• search_region : Whole brain
• statistic_type : For the group analysis we performed cluster-wise inference using a predefined cluster-forming threshold of p<0.001 (SnPM 'fast' option), cluster size was automatically estimated by SnPM (Statistical non Parametric Mapping, version 13.1.06)
• pval_computation : Permutation based p-values were calculated using SnPM. For estimation we used 15000 permutations. Variance smoothing was not applied.
• multiple_testing_correction : We used FWE correction p<0.05 on the cluster level. Details are specified above.
• comments_analysis : NA

Categorized for analysis

• region_definition_vmpfc : neurosynth, visually
• region_definition_striatum : neurosynth, visually
• region_definition_amygdala : neurosynth, visually
• analysis_SW : SPM
• analysis_SW_with_version : SPM12
• smoothing_coef : 8
• testing : permutations
• testing_thresh : p<0.001
• correction_method : GRTFWE cluster
• correction_thresh_ : p<0.05

Derived

• n_participants : 106
• excluded_participants : 003, 107
• func_fwhm : 8
• con_fwhm :

Comments

• excluded_from_narps_analysis : No
• exclusion_comment : N/A
• reproducibility : 3
• reproducibility_comment :

[ebannier]

Hi, Jumping in with the idea to contribute with a batch of the individual analysis

[cmaumet]

I am adding @ebannier as assignee (just for the time of the hackathon). So that we can more easily which pipelines are open for contributions in see this view: https://github.com/orgs/Inria-Empenn/projects/1/views/1

[bclenet]

Hi @cmaumet, I noticed that this pipeline uses SnPM :

independent_vars_higher_level : For hypothesis 1 to 8 first level contrasts were submitted to a one-sample permutation t-test (SnPM).
[...]
statistic_type : For the group analysis we performed cluster-wise inference using a predefined cluster-forming threshold of p<0.001 (SnPM ‘fast’ option), cluster size was automatically estimated by SnPM (Statistical non Parametric Mapping, version 13.1.06)

I think we should discuss about how to integrate SnPM in the project... Thanks !

[cmaumet]

We can do this without SnPM up to the production of the statistic map and then decide if we threshold using parametric statistics (as usual, in departure from what was done by the team here) or if we install SnPM to get the thresholded map using non-parametric computation of p-values.

[bclenet]

Code works with 4 subjects, to be tested with108.

[bclenet]

Correlation results with 106 subjects (sub-003 and sub-107 excluded).
[0.76, 0.96, 0.76, 0.96, 0.71, 0.71, 0.71, 0.71, 0.87]