Back

Unsupervised seizure annotation and detection with neural dynamic divergence

Ojemann, W. K. S.; Xu, Z.; Shi, H.; Walsh, K.; Pattnaik, A. R.; Sinha, N.; Lavelle, S.; Aguila, C.; Gallagher, R.; Revell, A. Y.; LaRocque, J. J.; Korzun, J.; Kulick-Soper, C. V.; Zhou, D. J.; Galer, P. D.; Sinha, S. R.; Shinohara, R.; Davis, K. A.; Litt, B.; Conrad, E. C.

2026-02-17 neurology
10.64898/2026.02.15.26346325 medRxiv
Show abstract

Annotating seizure onset and spread in intracranial EEG is essential for epilepsy surgical planning, yet manual annotation is unreliable and cannot scale to large datasets. We introduce Neural Dynamic Divergence (NDD), an unsupervised framework that detects seizure activity by measuring deviation from patient-specific baseline neural dynamics using autoregressive models. NDD requires no labeled training data and adapts to individual patients, channels, and brain states. Validating against expert consensus annotations from 46 seizures, NDD achieves human-level agreement ({phi} = 0.58 vs. inter-rater{phi} = 0.64) and outperforms existing algorithms on 1,019 seizures with soft labels (AUROC = 0.87). We demonstrate clinical utility by automatically annotating 2,017 seizures, revealing that seizure spread patterns distinguish epilepsy subtypes and predict surgical outcomes. NDD also generalizes to continuous ICU scalp EEG monitoring (AUROC = 0.77). We provide NDD as an open-source Python package to enable scalable seizure annotation across research centers.

Matching journals

The top 7 journals account for 50% of the predicted probability mass.

1
Nature Communications
○ 5641 papers in training set
Top 15%
12.1%
Similar papers in this journal
2
Brain
◐ 168 papers in training set
Top 0.4%
9.5%
Similar papers in this journal
3
Communications Medicine
○ 113 papers in training set
Top 0.1%
7.6%
Similar papers in this journal
4
Nature Computational Science
○ 55 papers in training set
Top 0.1%
7.0%
Similar papers in this journal
5
Nature Medicine
○ 125 papers in training set
Top 0.2%
6.5%
Similar papers in this journal
6
Epilepsia
○ 56 papers in training set
Top 0.2%
6.1%
7
Nature Biomedical Engineering
○ 47 papers in training set
Top 0.2%
5.3%
50% of probability mass above
8
Nature Neuroscience
○ 252 papers in training set
Top 2%
4.7%
9
Brain Communications
◐ 166 papers in training set
Top 1%
3.9%
10
Advanced Science
○ 286 papers in training set
Top 3%
3.1%
11
Proceedings of the National Academy of Sciences
● 2444 papers in training set
Top 19%
2.7%
12
npj Digital Medicine
○ 118 papers in training set
Top 2%
2.6%
13
Annals of Neurology
○ 64 papers in training set
Top 0.8%
2.3%
14
Scientific Reports
○ 3612 papers in training set
Top 43%
2.3%
15
Nature
○ 645 papers in training set
Top 6%
2.1%
16
Nature Machine Intelligence
○ 70 papers in training set
Top 1%
1.8%
17
Neuron
○ 337 papers in training set
Top 4%
1.6%
18
Communications Biology
○ 993 papers in training set
Top 16%
1.6%
19
eLife
● 5828 papers in training set
Top 57%
1.3%
20
Nature Cancer
○ 39 papers in training set
Top 1%
1.1%
21
The Journal of Neuroscience
● 1025 papers in training set
Top 9%
1.0%
22
Journal of Neural Engineering
○ 221 papers in training set
Top 2%
1.0%
23
PLOS Computational Biology
● 1863 papers in training set
Top 20%
0.8%
24
Med
○ 39 papers in training set
Top 0.8%
0.8%
25
iScience
○ 1154 papers in training set
Top 38%
0.8%
26
Nature Genetics
○ 286 papers in training set
Top 5%
0.8%
27
npj Parkinson's Disease
○ 105 papers in training set
Top 1%
0.8%

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.