LICENSE-THIRD-PARTY.md

Third-party notices

cortex-score bundles or invokes third-party assets under their own licenses. The MIT license in LICENSE covers the cortex-score source code only.

TRIBE v2 (Meta FAIR)

• Upstream: https://github.com/facebookresearch/tribev2
• Model card: https://huggingface.co/facebook/tribev2
• License: CC-BY-NC-4.0 (Creative Commons Attribution-NonCommercial 4.0)

The full score() path of this library loads and runs TRIBE v2. Any output produced by that path therefore inherits the upstream license restrictions, which prohibit commercial use without separate permission from Meta. Review the Hugging Face model card and the CC-BY-NC-4.0 license before using cortex-score in a commercial product.

TRIBE v2 also requires access to the gated meta-llama/Llama-3.2-3B text encoder; users must accept the LLaMA license on Hugging Face and run huggingface-cli login before the full inference path will work.

The CPU-only score_from_predictions / score_from_prediction_bundle paths of this library do not load TRIBE v2 — those paths receive an already-computed prediction tensor and only perform aggregation, normalization, metrics, and JSON serialization. They still operate on TRIBE-derived outputs in normal use, so the CC-BY-NC-4.0 attribution still applies to the resulting scores, but the dependency surface stays light.

Schaefer 2018 cortical parcellation

• Reference: Schaefer et al., Cerebral Cortex 28(9): 3095-3114 (2018). "Local-Global Parcellation of the Human Cerebral Cortex from Intrinsic Functional Connectivity MRI."
• Source: https://github.com/ThomasYeoLab/CBIG/tree/master/stable_projects/brain_parcellation/Schaefer2018_LocalGlobal
• License: MIT (CBIG repository).

The schaefer400_vertex.npy and labels_schaefer400.json files in cortex_score/data/ are derived from the published 400-parcel / 17-network projection on the fsaverage5 cortical mesh.

Yeo 2011 functional networks

• Reference: Yeo et al., Journal of Neurophysiology 106(3): 1125-1165 (2011). "The organization of the human cerebral cortex estimated by intrinsic functional connectivity."
• Source: https://surfer.nmr.mgh.harvard.edu/fswiki/CorticalParcellation_Yeo2011
• License: FreeSurfer license (BSD-like).

The yeo17_vertex.npy and labels_yeo17.json files in cortex_score/data/ are derived from the 17-network solution projected to fsaverage5.

5-network rollup (network_groups.json)

• Source: Cortexia project (packages/shared-schemas/src/clipcortex_schemas/network_groups.json).
• License: MIT (same as cortex-score).
• Group source identifier: cortexia-network-groups-v1.

The mapping of 17 Yeo networks into 5 dashboard groups (visual, language, faces, attention, motion) is a product-design choice of the Cortexia project, not a canonical neuroscience decomposition. The file is auditable: each group records its yeo_indices and the bundle ships a SHA-256 fingerprint in data/manifest.json.

Downstream attribution

If you publish or distribute results computed with cortex-score, please cite:

• Cite TRIBE v2 (see https://huggingface.co/facebook/tribev2 for the authoritative citation BibTeX).
• Cite Schaefer 2018 and Yeo 2011 if you report ROI- or network-level values.
• Cite cortex-score itself via the entry in CITATION.cff.