Open Neuroscience Infrastructure for Intracranial EEG

A toolbelt, not one solution — and an invitation to help bridge academia and the clinic

Yaroslav O. Halchenko @yarikoptic @yarikoptic@fosstodon.org	Cody Baker cody.c.baker.phd@gmail.com ORCID 0000-0002-0829-4790
Center for Open Neuroscience Department of Psychological and Brain Sciences Center for Cognitive Neuroscience Dartmouth College, New Hampshire, USA

Dartmouth/DH Human Intracranial Electrophysiology Consortium — October 9th, 2026
Live slides/Sources: https://datasets.datalad.org/centerforopenneuroscience/talks/2026-dhmc-ieeg.html

Before we start

What does your pipeline look like today?

acquisition → preprocessing → analysis → sharing (or not)

(~5 min — we'll shape the rest of the talk around what you tell us)

One toolbelt, not one solution

"Many sizes fit all" — no single tool or standard solves everything
You don't have to "buy in 100%" — pieces compose, and are useful alone
Everything today is a pointer: a name, a URL, a reason to look closer — not a deep technical tutorial

Where this is going

Two worlds that talk past each other — and a group sitting exactly in the middle

graph LR
  A[Academia / open neuroscience] -. weak .-> B(DHMC iEEG group - already here) -. weak .-> C[Clinic / hardware / vendors]
  classDef aca fill:#dff0d8,stroke:#3c763d;
  classDef bridge fill:#fcf8e3,stroke:#8a6d3b;
  classDef clin fill:#d9edf7,stroke:#31708f;
  class A aca
  class B bridge
  class C clin

BIDS / HED, NWB, DataLad,
OpenNeuro / NEMAR / DANDI

DICOM WG-16 ↔ BIDS, neuroconv, OBC (consent)

Natus / Blackrock / Ripple /
Nihon Kohden / Medtronic, DICOM, PACS, EHR/EMR

Carrots, not (only) sticks

Figure from M. Hanke, "Carrots! Not sticks!" hedgedoc.psychoinformatics.de/p/gBE63_ytiE

What pulls people toward standards

Carrots — the standards themselves: free tools, free archive storage, "someone else already solved this", collaborators who can actually read your data
Sticks exist too — mostly from funders and journals: NIH Data Management & Sharing Policy, journal data-availability requirements
Our pitch today is almost entirely carrots — the stick is already pulling on everyone; we're here to make the carrot side concrete

## Standards as an enabler for tooling

"Cool stuff (visualization, turnkey apps) is possible only because of boring stuff (standards)"

----

### Why do we need standards?

> Researcher A collects data, publishes a paper. Researcher B requests the data.
> The data arrive — organized however Researcher A felt like organizing them.
> Researcher B gives up.

<small>adapted from Rokem/Gau/Halchenko, <code>202404xx-NSF-POSE-BIDS</code></small>

----

### BIDS: Brain Imaging Data Structure

<small>
  <a href="https://www.nature.com/articles/sdata201644">Gorgolewski, K. J. et al. (incl. <b>Halchenko, Y. O.</b>) (2016).
  The brain imaging data structure... Scientific Data, 3:160044</a>
</small>

- a standard for organizing and describing neuroscience data
- a set of tools and services that use it (validators, BIDS-Apps, PyBIDS, ...)
- a community that develops, maintains, and extends it — **200,000+ individuals'** data stored in BIDS today

----

### Modular, like separating pasta ingredients from the dish

- **sourcedata/** — raw ingredients, kept as-is
- **derivatives/** — each processing pipeline's output, separate and swappable
- "you've seen one BIDS dataset, you've seen them all" — `sub-<label>/[ses-<label>/]<datatype>/`

<small>analogy via Rémi Gau, <a href="https://doi.org/10.5281/zenodo.5863713">10.5281/zenodo.5863713</a></small>

----

### iEEG is a first-class BIDS citizen

- **BEP010** — intracranial EEG extension
  (<a href="https://doi.org/10.1038/s41597-019-0105-7">Holdgraf et al., Scientific Data, 2019</a>)
- `channels.tsv` — per-channel metadata (type, sampling frequency, status)
- `electrodes.tsv` — electrode coordinates; clarified **REQUIRED** for iEEG in
  <a href="https://github.com/bids-standard/bids-specification/pull/1896">bids-specification PR#1896</a> (v1.10.1)
- **HED** (Hierarchical Event Descriptor) tags for event annotation — see NEMAR, next section

----

### Inheritance: don't repeat 320 JSON files

40 subjects × 2 sessions × 4 runs = 320 files to edit? No —
one `task-xyz_ieeg.json` at the dataset root covers all of them unless overridden.

NWB: one file format for all modalities + metadata

neuroconv.readthedocs.io

Neurosift: stream, don't download

Browser-based NWB viewer — no download required, works directly against DANDI
Open-source
New event-related signals view: align any time series to any column of a trials/intervals table, pick channels, group trials by a condition

One known iEEG NWB example on DANDI: ds008610 (LFP only) — TODO: find/convert a second example with spectrograms if one exists.

Example 1: behavior, grouped by condition

Camera motion energy, aligned to auditory_cue_time, grouped by gabor_stimulus_contrast — DANDI:000409

Example 2: LFP aligned to reward

LFP aligned to reward_start_time, channels 0, 2, 4 — DANDI:001539

Example 3: LFP around stimulus presentations

40 Hz pulse-train presentations, probe D LFP, channels 5, 22, 25, 32 — DANDI:001637

## Archives: free backup, open by design

----

### NEMAR — Neuroelectromagnetic Data Archive and Tools Resource

[nemar.org](https://nemar.org) — SCCN + SDSC, UC San Diego (NIH/NIMH R24MH120037)

----

### NEMAR, verified facts (as of 2026-10-05)

- Covers **EEG, MEG, iEEG, EMG** — iEEG is first-class, not an afterthought
- Auto-mirrors new public OpenNeuro EEG/MEG/iEEG datasets within ~24h, BIDS-formatted
- **762 datasets, ~40K participants, 59.2 TB** hosted today
- Built-in **HED** annotation support + search/filter by HED coverage
- Compute bridge to SDSC / Neuroscience Gateway — analyze without downloading
- An experimental "NEMAR Assistant" chatbot answers questions and plots recordings
  (general-purpose today, not a dedicated HED-annotation generator —
  that piece is still aspirational)

----

### DANDI — the broader NWB-centric archive

[dandiarchive.org](https://dandiarchive.org) — also hosts iEEG data in NWB form (see ds008610 above)

----

### EMBER — new since your notes were written

Official BRAIN Initiative archive for the **BBQS** program (JHU/APL-led, with **Dartmouth/CON**),
live since 2024, built on DANDI infrastructure. [emberarchive.org](https://emberarchive.org)

## RDM: Data Management

----

### `datalad run` — version control for outcomes, not just code

```
datalad run -m "run preprocessing" \
  "python preprocess.py {inputs} {outputs}"
```

- captures exact command, inputs, outputs, environment as a git commit
- `datalad rerun` replays it later, on different data, to check it still works
- works over arbitrarily large (binary) data via `git-annex`

----

### STAMPED — one slide

**S**elf-contained, **T**racked, **A**ctionable, **M**odular, **P**ortable, **E**phemeral, **D**istributable

<small><a href="https://stamped-principles.org/">stamped-principles.org</a> —
Macdonald, Baker, To, Halchenko (2026); Cody's deeper dive: BBQS workshop slides (<code>2026-bbqs-stamped</code>)</small>

----

### (Meta)BIDS-Study: self-containment + modularity together

<a href="https://github.com/OpenNeuroStudies/OpenNeuroStudies">OpenNeuroStudies</a> —
a dataset of datasets, each still independently valid BIDS

Compute: pointers only

ReproNim/containers — portable, versioned software environments
mechababs + BABS — STAMPED execution of BIDS-Apps at scale, DataLad-native, per-subject isolation
DANDI Compute — standardized processing of NWB data on DANDI (next slides)
TODO: "compute on Unity" — need one line from you/Austin; nothing in our existing materials mentions this

DANDI Compute

DANDI Compute: orchestration

DANDI Compute: the spike-sorting pipeline

DANDI Compute: the derivative it produces

## Visions for the future

----

### Mass preprocessed data, reused with confidence

- "Mass compute" (see mechababs) makes mass-preprocessing tractable
- Goal: future "meta-studies" / multiverse analyses that trust someone else's preprocessing
  because STAMPED provenance makes it checkable

----

### Knowledge bases: from local to global

<div style="display:flex;justify-content:space-around;align-items:flex-start;">
<div><img data-src="pics/borrowed/brainkb-webshot-20261006.png" style="width:440px;"/><br/>
  <small><a href="https://brainkb.org">BrainKB</a> — NIH-funded open neuroscience knowledge graph
  (BICAN, ReproNim, BBQS, Brain Connects)</small></div>
<div><img data-src="pics/borrowed/ebrains-kg-webshot-20261006.png" style="width:440px;"/><br/>
  <small><a href="https://ebrains.eu">EBRAINS Knowledge Graph</a> — EU openMINDS-based
  semantic metadata graph</small></div>
</div>

----

### ORINOCO / ORINOCO-Lite — our own tooling, not an external project

- `dump-things-service` — self-hostable LinkML-based structured-metadata store
- `ORINOCO-Lite` — GitHub-native deployments, already running for CON and
  Psychoinformatics group sites
- Local knowledge bases that could, in principle, federate upward into
  BrainKB / EBRAINS-scale graphs

All standards are "Bad"^*,
but some are used

* ref: D. Clunie, MICCAI 2017

Two widely-used standards which could cooperate more

BIDS (academia) and DICOM (clinic/vendors) — same underlying problem, different communities, thin collaboration today

see 2023-bids-dicom.html (BIDS 4 DICOM WG-16 talk) for the longer version of this argument

How could we collaborate on metadata that matters clinically?

Should individual labs channel requests to manufacturers one at a time, or could DICOM Working Groups (e.g. WG-16) help coordinate?
Proposition: a library of concrete DICOM↔BIDS/NWB conversions, per manufacturer and modality

You're already standing in the gap

This group runs clinical iEEG acquisition and does research analysis — both worlds, already, every day
That makes you natural informants for neuroconv format support, BEP010 edge cases, and DICOM↔BIDS/NWB conversion libraries
Ask: who else should be in that room? what hardware/format gaps have you hit that we should know about?

Acknowledgements

Collaborators & sources

Ariel Rokem, Rémi Gau — BIDS-101 slide material
Michael Hanke — "Carrots! Not sticks!" figure
Austin Macdonald — mechababs
SCCN/NEMAR, EMBER/BBQS, BrainKB, EBRAINS teams

Funders

Collaborators

Open Neuroscience Infrastructure for Intracranial EEG

Before we start

One toolbelt, not one solution

Where this is going

Carrots, not (only) sticks

What pulls people toward standards

NWB: one file format for all modalities + metadata

Neurosift: stream, don't download