Post-Processing

This page describes outputs of the post-processing utilities ModelArray and ReproTM, generated from ABCC pipeline derivatives and included in the release.

ModelArrayIO

Mass-univariate statistical modeling for large neuroimaging datasets

ModelArrayIO is a Python package that converts neuroimaging file formats to HDF5 (.h5) for compatibility with the ModelArray R package. For ABCC, ModelArrayIO is used as a downstream aggregation step, converting subject-level outputs from QSIRecon and XCP-D into cohort-level arrays. This enables efficient, large-scale statistical analyses via ModelArray without having to load individual subject files.

Note: Each .h5 file is paired with a .csv file indexing included subjects and sessions.

Outputs were generated via the following command (see Usage Notes for details):

modelarrayio to-modelarray \
    --cohort-file ${cohort_csv} \
    --output ${out_h5} \
    --scalar-columns ${SCALAR_COLUMNS} \
    --backend hdf5 \
    --dtype float32 \
    --compression gzip \
    --compression-level 9 \
    --s3-workers 4 \
    --log-level INFO \
    --chunk-voxels 0 \
    --target-chunk-mb 32

QSIRecon ModelArray

QSIRecon-ModelArray outputs include cohort-level HDF5 arrays aggregating diffusion MRI scalar maps across QSIRecon reconstruction methods, including DIPY, DSI Studio, TORTOISE, and wmNODDI (see QSIRecon for details).

abcd/
└── derivatives/
    └── abcc-qsirecon-ModelArray/

        # DIPY outputs
        ├── qsirecon-DIPY_model-<tensor|dki>_param-<PARAM>.h5
        ├── qsirecon-DIPY_model-<tensor|dki>_param-<PARAM>.csv

        # DSI Studio outputs
        ├── qsirecon-DSIStudio_model-<tensor|gqi|rdi>_param-<PARAM>.h5
        ├── qsirecon-DSIStudio_model-<tensor|gqi|rdi>_param-<PARAM>.csv

        # TORTOISE outputs
        ├── qsirecon-TORTOISE_model-tensor_param-<PARAM>.h5
        ├── qsirecon-TORTOISE_model-tensor_param-<PARAM>.csv

        # wmNODDI outputs
        ├── qsirecon-wmNODDI_model-noddi_param-<PARAM>.h5
        └── qsirecon-wmNODDI_model-noddi_param-<PARAM>.csv

XCP-D ModelArray

XCP-D-ModelArray outputs include cohort-level HDF5 arrays aggregating structural and functional derivatives, including:

  • [1] Surface morphometry (curvature, sulcal depth, cortical thickness)
  • [2] Functional maps (ALFF, ReHo)
  • [3] Resting-state functional connectivity.<ATLAS> = Gordon, HCP, MIDB (see details)
abcd/
└── derivatives/
    └── abcc-xcp_d_v0.13.0-ModelArray/

    # Surface morphometry [1]
        ├── xcp_d_v0.13.0_param-<curv|sulc|thickness>.h5
        ├── xcp_d_v0.13.0_param-<curv|sulc|thickness>.csv

    # Functional maps [2]
        ├── xcp_d_v0.13.0_param-<alff|reho>.h5
        ├── xcp_d_v0.13.0_param-<alff|reho>.csv

    # Resting-state functional connectivity [3]
        ├── xcp_d_v0.13.0_task-rest_seg-<ATLAS>_param-pearsoncorrelation.h5
        └── xcp_d_v0.13.0_task-rest_seg-<ATLAS>_param-pearsoncorrelation.csv

ReproTM

Reproducible Template Matching (ReproTM) (Godfrey, 2026) is a Python pipeline that generates individualized (i.e., subject-specific) functional network maps from dense functional connectivity data using the template matching network detection algorithm (Gordon et al., 2017; Hermosillo et al., 2024).

ReproTM compares the functional connectivity profile of each greyordinate against a set of canonical resting-state functional network templates to produce subject-specific functional network assignments. The primary output is a labeled CIFTI (.dlabel.nii) file assigning each greyordinate to one of 15 functional networks defined within the ABCC 2026 template.

ABCC 2026 Template Networks

The ABCC 2026 templates were derived from ABCC Group 3 (n = 561 subjects) to capture canonical functional network patterns optimized for the developmental age range and scanning parameters of the ABCC cohort. The template set includes 15 canonical functional resting state networks:

NoteABCC 2026 Functional Networks
[1] Default Mode
[2] Visual
[3] Frontoparietal
[4] Dorsal Attention
[5] Ventral Attention		 [6] Salience
[7] Action Mode
[8] Sensorimotor Dorsal
[9] Sensorimotor Lateral
[10] Auditory		 [11] Temporal Pole
[12] Medial Temporal Lobe
[13] Parietal Memory
[14] Parietal Occipital
[15] Somato-Cognitive Action

Outputs

ReproTM derivatives include greyordinate-wise network assignments (.dlabel.nii) and intermediate greyordinate-to-network similarity matrices and assignments (.mat) organized under abcc-ReproTM/:

abcc-ReproTM/
├── templates/
│   └── tpl-ABCC2026-a3-9to16_space-fsLR_den-91k_desc-seedmap_stat-zscored.mat
│
└── sub-<label>/
    └── ses-<label>/
        └── func/
            ├── *.dlabel.nii    # Final network assignments
            └── *.mat           # Intermediate metrics and assignments
  • The .dlabel.nii file contains greyordinate-wise functional network assignments derived from the preprocessed and denoised BOLD timeseries using the ABCC 2026 template networks.
  • The accompanying .mat file contains ReproTM intermediates from which the grayordinate-wise network assignments were derived, including:
    • Similarity values (eta and linear bivariate r) for each greyordinate to every template network
    • Network assignments pre- and post-SCAN network refinements
sub-<label>_ses-<label>_task-rest
  _space-fsLR_den-91k
  _desc-denoised-spatially-interpolated-smoothed-2.25mm-censor-ReproTM
  _template-ABCC-a3-9to16
  _refine-SCAN_minsize-30
  _boldmap.{dlabel.nii|mat}

Processing

Prior to running ReproTM, subject-level dense connectivity matrices (dconn.nii) were generated from XCP-D resting state timeserives derivatives using cifti-connectivity. ReproTM v1.0.0 was then applied to compare dense connectivity matrices to the ABCC 2026 group-level template file provided in the derivatives and in the ReproTM Github repository.

Cifti-Connectivity Workflow & Parameters

  1. Interpolate timeseries for zero-value greyordinates using the mean and standard deviation of greyordinates in the same structure.
  2. Spatial smoothing.
  3. Dense connectivity matrix generation with motion censoring.
  • Input data:
    • Denoised resting-state CIFTI timeseries (*_desc-denoised_bold.dtseries.nii)
    • Anatomical surface files in fsLR 32k space
    • Quality control outputs (*_desc-abcc_qc.hdf5)
  • Parameters:
    • Repetition time: 0.8 s
    • Motion threshold: FD < 0.2 mm
    • Minimum usable data: 10 minutes
    • Spatial smoothing: 2.25 mm FWHM

ReproTM Workflow & Parameters

  1. Z-score connectivity matrices across greyordinates
  2. Perform template matching with initial template network thresholding (z=1.00)
  3. Perform optional refinement of somatomotor and somato-cognitive action networks with higher template network thresholding (z=3.00)
  4. Apply minimum network size filtering (minimum 30 vertices)
  5. Convert outputs to labeled CIFTI format (.dlabel.nii)

Parameters are included in the run command used to generate ReproTM outputs for ABCC below:

python3 ReproTM_v1.0.0.py \
  ${zscored_dconn_file} \
  --template_infile \
    tpl-ABCC2026-a3-9to16_space-fsLR_den-91k_desc-seedmap_stat-zscored.mat \
  --template_networks \
    "DMN Vis FP NaN DAN NaN VAN Sal AMN \
     SMd SMl Aud Tpole MTL PMN PON NaN SCAN" \
  --template_thresholding \
  --template_minthreshold 1 \
  --refineSCAN \
  --refineSCAN_minthreshold 3

References

Godfrey, K. J. (2026). Zenodo. https://doi.org/10.5281/zenodo.20139438
Gordon, E. M., Laumann, T. O., Adeyemo, B., & Petersen, S. E. (2017). Cerebral Cortex, 27(1), 386–399. https://doi.org/10.1093/cercor/bhv239
Hermosillo, R. J. M., Moore, L. A., Feczko, E., Miranda-Domínguez, Ó., Pines, A., Dworetsky, A., Conan, G., Mooney, M. A., Randolph, A., Graham, A., Adeyemo, B., Earl, E., Perrone, A., Carrasco, C. M., Uriarte-Lopez, J., Snider, K., Doyle, O., Cordova, M., Koirala, S., … Fair, D. A. (2024). Nature Neuroscience, 27(5), 1000–1013. https://doi.org/10.1038/s41593-024-01596-5