BCI Report Research previewPeople
149
100 people with Parkinson's disease and 49 controls, one site
Clinical research · resting-state EEG
Clinical groups
On one clinic's resting recordings, a simple spectral model separated the two groups at 71.7% balanced accuracy, where chance is 50%. A model given only age and sex, and no EEG at all, reached 55.0% on the same folds. Both are research results on a public data set. Neither is a diagnosis, neither has been repeated on another cohort, and the comparator is printed here precisely because a case/control score alone does not say what produced it.
+16.6 ppEEG minus the age-and-sex comparator, same 149 people
The EEG model scores above a model with no EEG
71.7% against 55.0%, on the same folds and the same people. The descriptive intervals are 63.5%–79.3% and 46.9%–63.6%, and they overlap. One split seed, no tuning, one site: this is a fixed experiment, not an estimate of what such a model would do elsewhere.
Research results on a public data set. Not a diagnosis, not diagnostic accuracy, and not a medical device or medical advice.
Two rows, one of which is not a model
EEG against age and sex
Two classes, chance 50%. 5 stratified participant folds; every person is held out exactly once, and one averaged feature vector per person enters the classifier.
| Row | Inputs | Balanced accuracy | Macro F1 | AUROC |
|---|---|---|---|---|
| Relative-spectrum logistic regressionlog relative band power, 60 shared channels | EEG only | 71.7%63.5%–79.3%71.7% balanced accuracy | 71.1% | 0.76 |
| Age and sex onlynot an EEG model · excluded from every comparison | no EEG | 55.0%46.9%–63.6%55.0% balanced accuracy | 54.0% | 0.56 |
What the scores are computed on
One site, 149 people, two minutes each
The 60 named channels shared by four source layouts; the first 120 seconds of each recording, cut into 30 disjoint four-second windows; log relative power in five bands spanning 1-45 Hz. Each person's 300 features are averaged within that person, so the classifier sees one vector per person rather than 4,470 independent examples. Scaling is fitted on training folds only, the classifier is a fixed balanced logistic regression (C=1), and there is one split seed with no held-out tuning.
Participant folds
5
each person held out once
Four-second windows
4,470
4,470, averaged within each person
Methods & limits
What this result is not
A case/control comparison on one cohort, with one simple model and one split. The list below is not boilerplate: each line is a claim this number cannot support.
Not a diagnosis
This is group discrimination on a fixed research split, reported as balanced accuracy. It is not diagnostic accuracy, not a screening tool, not clinical validation, and it supports no interpretation about any individual person.
Not replicated
One site, one cohort, one split seed, no held-out tuning. The intervals are descriptive bootstrap ranges, not population guarantees — the cross-validation training folds overlap, which is exactly why they are called descriptive here.
Age and sex are not removed
The comparator shows that part of this separation is available with no brain signal at all. The EEG model is not adjusted for age or sex, so its score is not a measure of disease-specific brain activity.
Group differences are not only in the brain
Anything that differs between patients and controls travels with the labels: medication, recording session, movement, alertness. A case/control score cannot separate those from the disease.
No foundation-model score for this data set
The source does not state a physical amplitude unit. A relative spectrum does not need one, so this result stands; models that need a calibrated scale were left unrun rather than run on an assumed unit.
Separate from the matrix
This release is its own review and its own file. It does not join the eight-protocol matrix or the deployment topics, and it shares no ranking with them.
Independently replayed
149 recordings and 4470 spectral windows with a separate FFT, all classifiers refit. Reconstructed features and metrics differed by 0; intervals differed at floating-point precision.
OpenNeuro ds004584 · Rest eyes open, Parkinson's disease and controls
Dataset: Arun Singh, Rachel Cole, Arturo Espinoza, Jim Cavanagh and Nandakumar Narayanan · Rest eyes open, OpenNeuro ds004584 v1.0.0, doi:10.18112/openneuro.ds004584.v1.0.0. Cohort study: Arun Singh, Rachel C. Cole, Arturo I. Espinoza, Jan R. Wessel, James F. Cavanagh and Nandakumar S. Narayanan · Evoked mid-frontal activity predicts cognitive dysfunction in Parkinson's disease, Journal of Neurology, Neurosurgery & Psychiatry 94, 945-953 (2023), doi:10.1136/jnnp-2022-330154. Transport mirror: NEMAR on004584, doi:10.82901/nemar.on004584.
Every participant gave written informed consent, with decisional capacity established, in accordance with the Declaration of Helsinki; all research protocols were approved by the University of Iowa Human Subjects Review Board (IRB 201707828). Only cohort aggregates appear here: no participant rows, no clinical or cognitive scores, and no per-group age or sex composition.
OpenNeuro dataset ↗ · Cohort study ↗ · Open manuscript ↗ · NEMAR mirror ↗ · CC0-1.0
Data source: reviewed aggregate JSON · schema bci-report-clinical-update-v1 · generated 2026-09-23.