Lorente Lemoine, Loic, Huynh, Kha, Beuchert, Jonas, John, Deepu, Rana, Omer  ORCID: https://orcid.org/0000-0003-3597-2646 and Pham, Nhat (Nick)
2026.
PERSE: A PERsonalised SEizure prediction system based on large-scale EEG foundation models.
Presented at: MobiSys Workshop '26: 24th Annual International Conference on Mobile Systems, Applications and Services Workshops,
Cambridge, UK,
21-25 June 2026.
Proceedings of the 24th Annual International Conference on Mobile Systems, Applications and Services Workshops.
ACM,
pp. 81-86.
10.1145/3812836.3814759
|
Preview |
PDF
- Published Version
Available under License Creative Commons Attribution. Download (11MB) | Preview |
Abstract
Annually, epilepsy exacts a global economic toll exceeding $199 billion and impacts over 51 million individuals, resulting in 140,000 fatalities. Beyond the threat of early mortality, the condition significantly diminishes the quality of life for both patients and their families through its enduring long-term effects. There is an urgent need for an effective seizure prediction solution. Anticipating an onset would empower patients and providers to mitigate risks, while enabling closed-loop systems to deliver targeted interventions to suppress seizures. We proposed PERSE, an early seizure prediction system based on a large foundation model, pre-trained on 25,000 subjects and 60,000 hours of EEG data. Our preliminary evaluations on 10 epileptic patients show encouraging results. In intra-subject evaluation, PERSE achieves average scores of 0.80, 0.78, and 0.81 for sensitivity, specificity, and AUROC, i.e., Area Under the Receiver Operating Characteristic curve, respectively. On unseen subjects, PERSE achieves the average of 0.58, 0.58, and 0.60 in sensitivity, specificity and AUROC score. By combining general knowledge from the foundation model with minimal prior subjects' knowledge, we significantly improve the average sensitivity, specificity, and AUROC scores to 0.78, 0.78, and 0.81.
| Item Type: | Conference or Workshop Item - published (Paper) |
|---|---|
| Date Type: | Publication |
| Status: | Published |
| Schools: | Schools > Computer Science & Informatics |
| Additional Information: | License information from Publisher: LICENSE 1: URL: https://creativecommons.org/licenses/by/4.0/legalcode, Start Date: 2026-06-20 |
| Publisher: | ACM |
| ISBN: | 9798400727122 |
| Date of First Compliant Deposit: | 13 July 2026 |
| Last Modified: | 04 Aug 2026 23:47 |
| URI: | https://orca.cardiff.ac.uk/id/eprint/187634 |
Actions (repository staff only)
 |
Edit Item |
Dimensions
Dimensions