PMC open-access papers with fMRI in the title or abstract · 2016 full year vs 2026 to Oct 21

fMRI coordinate sharing, 2016 vs 2026

Coordinates or NeuroVault maps appear in — of 2026 fMRI research articles, down from — in 2016. Supplementary files add back only — (2026) and — (2016). The rest of the drop went to papers whose results are parcel-level maps or that produce no whole-brain map at all.

Coordinates or NeuroVault in the paper
—
2016: — · change —
Share of research articles, 2026
Coordinates or maps only in a supplement
—
2016: — · change —
Share of research articles, 2026
Parcel map only, not shared
—
2016: — · change —
Share of research articles, 2026
No whole-brain map
—
2016: — · change —
Share of research articles, 2026
Open-access fMRI papers
—
2016: — · about — for a full 2026
All article types, 2026 to Oct 21

Where fMRI research articles land

Every research article in each year's corpus. Papers the coordinate screen flagged, plus papers where coding found coordinates the screen missed, count as coordinates or NeuroVault in the paper. Papers whose only coordinates or brain maps sit in supplementary files form their own group. The rest are split using each year's coded sample.

What the coordinate screen found

Run on every downloaded paper. pubget reads coordinate tables and NeuroVault links; extra detectors add NeuroVault mentions, coordinates written in text or captions, and coordinate tables pubget could not parse. Rows are sorted by change in percentage points from 2016 to 2026.

20162026

Papers without coordinates, by what they produce

In-scope coded papers with no detected coordinates or NeuroVault link (— in 2016, — in 2026). Ineligible papers are — of them in 2016 and — in 2026. Lines show 95% intervals. Within each group, rows are sorted by change in percentage points from 2016 to 2026.

20162026

Eligible means a whole-brain voxel or parcel map is a result of the paper. A voxel map used only as an intermediate step, such as a localizer that defines ROIs, files the paper under its headline result in the ineligible group.

Inside the parcel-map bucket

Papers whose only whole-brain result is a value per atlas parcel: — coded papers in 2016 and — in 2026. Click a row to list its papers in the table below.

What these papers lead with

Share of all in-scope coded papers without coordinates, so the rows add up to the parcel-map bucket. Sorted by change in percentage points.

20162026

Traits of parcel-map papers

Share of the parcel-map papers in each year with each trait. A paper can have several. Sorted by change in percentage points; the 2016 bucket is small.

20162026

Coordinates in supplementary files

Supplementary files for every sampled paper, fetched from PMC's open-access bucket. Tables and text were extracted from Word, Excel, PDF, CSV and zip files and scanned for coordinate tables, and a reader confirmed each hit as result peaks or as input coordinates such as seeds and ROI centres.

Sampled papers that… 2016 2026

Counts out of each year's coded sample. Result peaks also appear in some papers that already had coordinates in the main text or are out of scope, so the in-scope row is the one that changes the buckets.

How the analyses changed

Analysis approach

Share of in-scope coded papers without coordinates using each approach. A paper can use several. Rows are sorted by change in percentage points from 2016 to 2026.

20162026

Method keywords across all papers

Share of all full-text research articles, with or without coordinates, that mention each term group at least three times. Rows are sorted by change in percentage points from 2016 to 2026.

20162026

Naturalistic stimuli

Movies and video, spoken narratives, music, games and VR. The coded-sample chart was confirmed by reading each flagged paper. The all-papers chart is a keyword count (two or more mentions in the full text), so treat it as a rough trend.

In coded papers without coordinates

Share of in-scope coded papers. Sorted by change in percentage points. Click a row to list its papers.

20162026

Across all research articles (keyword count)

Share of full-text research articles mentioning naturalistic stimuli, by whether the paper reports coordinates or NeuroVault. Sorted by change in percentage points.

20162026

Where the naturalistic papers land

Coded papers without main-text coordinates that use naturalistic stimuli, by group.

Group20162026

Where the data came from and what was shared

Data source

In-scope coded papers without coordinates.

Most open thing shared

In-scope coded papers without coordinates, by data or code availability statement.

Public data sources

Which public datasets the coded papers without coordinates analyse, read and confirmed for every paper coded as using public data or mentioning a dataset. Atlases, templates and pipelines named after a dataset (such as the Glasser HCP-MMP atlas) do not count. A paper can use several datasets.

Datasets analysed

Share of in-scope coded papers. Sorted by change in percentage points. Click a named dataset to list its papers.

20162026

How the public data were used

Share of in-scope coded papers. Precomputed features are values a consortium released (such as ABCD tabulated connectivity), so the authors never had the images.

20162026

Dataset mentions across all research articles (keyword count)

Share of full-text research articles naming each dataset at least twice. A mention is not proof of use. Sorted by change in percentage points.

20162026

Browse the coded papers

Every paper in both samples with its bucket, headline result, the coder's own label and a one-line summary. Search matches any column.

How this was measured

Steps

  1. Downloaded every PMC open-access paper with fMRI in its title or abstract for each year, using pubget.
  2. Flagged papers with coordinate tables, coordinates in text or captions, coordinate tables pubget could not parse, or a NeuroVault link or mention.
  3. Drew — random research articles per year from the rest (— in 2016, — in 2026) and coded each from its abstract, methods, figure and table captions and data statement.
  4. Coders recorded facts such as the headline result and whether a whole-brain map exists. Buckets come from fixed rules applied to those facts.
  5. Fetched each sampled paper's supplementary files, scanned their tables and text for coordinates and brain-image files, and confirmed each coordinate hit by reading it. Verified result peaks or maps move a paper to eligible, shared.
  6. Read every in-scope coded paper flagged as naturalistic or as using public data, by its coding or by a keyword scan, to confirm the stimuli and name the datasets it analyses.
  7. Coded — papers per year a second time, independently, to check agreement.

Agreement between independent codings

Cohen's kappa and raw agreement. Edge scope in 2026 was added to both codings by one pass, so it has no agreement figure.

Field Kappa 2016 Agree 2016 Kappa 2026 Agree 2026

Caveats

  • The 2016 papers more often give a few bare peaks in captions or methods text that the screen misses (— of the 2016 coded sample). They count as coordinates here, which slightly widens the drop.
  • Coordinates used as inputs, such as seed centres and stimulation targets, count as detected in both years. That group has not been audited.
  • Supplements that are only images, such as scanned PDFs or tables pasted as pictures, could not be read. — sampled papers in 2016 and — in 2026 promise supplement coordinates that could not be read, and they stay in their earlier bucket.
  • The coders were one language model reading excerpts, not full papers or supplements. Agreement shows the rules are applied consistently, not that they match a human expert.
  • The open-access corpus changed: preprints are — of 2026 research articles and none of 2016's. Journal articles alone still go from — to — with coordinates or NeuroVault detected.
  • Coordinate-reporting papers still grew in number, from — in 2016 to about — for a full 2026. The decline is in share.