Neurosynth maps

Maps and topic-model outputs derived from the Neurosynth automated meta-analytic database. The headliner here is the v4 100-topic LDA topic model (in the 2016_Neurosynth_100_topics/ sub-folder), but the folder also holds several auxiliary resources: MKDA outputs, topic embeddings, co-activation matrices, and helper scripts.

Sub-folders

Folder	Description
2016_Neurosynth_100_topics	Pilot — Neurosynth v4 100-topic LDA model: 50 topics × 2 inference directions (pAgF + pFgA) at FDR q<0.01, plus combined fmri_data objects and GPT-5–derived topic labels.
mkda	MKDA per-term meta-analytic kernel-density maps from the Neurosynth corpus.
scripts	Helper scripts used to build and re-derive the above.
topic_terms_csv	Per-topic term-frequency CSVs.

Auxiliary files at this level

File	What it is
neurosynth_data_obj.mat	Aggregate fmri_data object covering the Neurosynth term/topic maps.
neurosynth_topics_grouped_by_clique.mat	Topics grouped by maximal-clique structure.
neurosynth_interregion_coactivation_canlab2018_2mm.mat	Inter-region co-activation matrix at 2 mm using the 2018 CANlab atlas.
canlab_combined_atlas_2018_parcel_means_neurosynth.mat	Parcel-mean Neurosynth maps under the 2018 combined atlas.
canlab_combined_atlas_2018_resorted_brainpathway_obj.mat	Brain-pathway object derived from those parcel means.
topic_embeddings_text-embedding-3-large.{csv,mat}	Per-topic vector embeddings (OpenAI text-embedding-3-large).
current_data.tar.gz	Compressed snapshot of the raw Neurosynth data used.
assign_unique_cliques_and_average.m, assign_unique_cliques_from_maximal_cliques.m, maximalCliques.m	Clique-finding helpers used to group similar topics.
neurosynth_interregion_coactivation.m, neurosynth_seed_coactivation_map.m	Build / use co-activation matrices.
2026_Neurosynth_topics_LLM_vector_embedding.pptx	Slide deck describing the LLM-embedding work.

Loading

The topic maps are registered in load_image_set.m:

[obj, ~, ~] = load_image_set('neurosynth_topics_forwardinference');
[obj, ~, ~] = load_image_set('neurosynth_topics_reverseinference');
[obj, ~, ~] = load_image_set('neurosynth');     % original term-level

To annotate an arbitrary brain map with its top Neurosynth topic:

new_data = fmri_data('my_contrast.nii');
[~, topics] = neurosynth_feature_labels(new_data, 'topics_ri');
disp(topics{1}.Term_or_Topic_highest(1:5));

Conventions

See the docs README.