pith. sign in

arxiv: 2606.22657 · v1 · pith:LHVSKME2new · submitted 2026-06-21 · 💻 cs.LG · cs.AI

Foundation Models for Epileptogenic Zone Identification in Drug-Resistant Epilepsy

Thi Kieu Khanh Ho , Thomas Lai , Petr Klimes , Jan Cimbalnik , Martin Pail , Milan Brazdil , Birgit Frauscher , Narges Armanfard This is my paper

Pith reviewed 2026-06-26 10:29 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords epileptogenic zonestereo-electroencephalographyfoundation modelsdrug-resistant epilepsyinterictal signalsleave-one-patient-outseizure onset zone
0
0 comments X

The pith

EpiiSLM identifies the epileptogenic zone at 0.978 contact-level PPV by training a signal foundation model on all sEEG recordings and anchoring biomarkers on non-epileptic signals.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents EpiiSLM, a dual foundation model system that first trains a signal model on 104,990 minutes of stereo-electroencephalography data from all patients irrespective of surgical outcome, then uses a language model to fuse the resulting biomarkers with clinical information. This produces predictions for the epileptogenic zone that reach 0.978 contact-level positive predictive value and 100 percent region-level accuracy under leave-one-patient-out testing while requiring only one night of interictal sleep recordings. The approach is tested on an external dataset where it attains 0.857 contact-level PPV and is shown to exceed the seizure-onset-zone baseline. A sympathetic reader would care because current resective surgery succeeds in fewer than half of drug-resistant epilepsy cases, so a more accurate, shorter-duration EZ marker could raise seizure-freedom rates.

Core claim

EpiiSLM achieves 0.978 contact-level positive predictive value and 100 percent region-level accuracy for epileptogenic-zone identification by training a signal foundation model on every available sEEG recording and anchoring biomarker extraction to non-epileptic signals, then feeding the outputs together with multimodal clinical data into a language foundation model; the same system yields 0.857 contact-level PPV on held-out external data while using only one night of interictal sleep recordings.

What carries the argument

Dual foundation model pipeline in which a signal foundation model extracts biomarkers from sEEG anchored on non-epileptic segments and a language foundation model integrates those outputs with clinical variables to produce EZ predictions.

If this is right

  • Only one night of interictal sleep sEEG is needed for high-accuracy EZ mapping.
  • Performance exceeds the seizure-onset-zone-as-EZ baseline by 15.1 percent at contact level.
  • The system maintains 0.857 contact-level PPV on an external validation set.
  • Region-level accuracy reaches 100 percent under leave-one-patient-out evaluation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Shortening sEEG monitoring to a single night could lower infection risk and hospital stay length for patients.
  • The same anchoring-on-non-epileptic-signals strategy might transfer to other invasive electrophysiological biomarkers.
  • Multimodal fusion of signal and language models could be tested on non-epilepsy sEEG or intracranial EEG datasets.

Load-bearing premise

Training the signal foundation model on recordings from every patient regardless of surgical outcome and anchoring biomarker extraction on non-epileptic signals produces a reliable, generalizable marker for the epileptogenic zone.

What would settle it

A prospective patient cohort in which EpiiSLM contact-level PPV falls below the seizure-onset-zone baseline or region-level accuracy drops below 100 percent under the same leave-one-patient-out protocol.

read the original abstract

Accurate identification of the epileptogenic zone (EZ) is essential for seizure freedom after resective surgery in drug-resistant epilepsy, yet seizure freedom rates remain below 50%. We developed EpiiSLM, a dual foundation model system for EZ identification with stereo-electroencephalography (sEEG), by training a signal foundation model on 104,990 minutes of sEEG recordings from the Montreal Neurological Institute & Hospital, while leveraging all recordings regardless of surgical outcome and anchoring EZ biomarker extraction on non-epileptic signals. A language foundation model then integrates sEEG-derived outputs with multimodal clinical information to produce interpretable predictions. Under leave-one-patient-out evaluation, EpiiSLM achieved 0.978 contact-level positive predictive value (PPV), outperforming the seizure onset zone(SOZ)-as-EZ baseline by 15.1% (p < 0.05), and 100% region-level accuracy; on an external dataset, EpiiSLM achieved 0.857 contact-level PPV. EpiiSLM requires only one night of interictal sleep data, suggesting potential to reduce invasive sEEG monitoring duration and improve surgical outcomes.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper introduces EpiiSLM, a dual foundation model system for epileptogenic zone (EZ) identification from stereo-electroencephalography (sEEG) in drug-resistant epilepsy. A signal foundation model is trained on 104,990 minutes of sEEG recordings from the Montreal Neurological Institute & Hospital, using all recordings irrespective of surgical outcome and anchoring biomarker extraction on non-epileptic signals; a language foundation model then integrates the outputs with multimodal clinical data. Under leave-one-patient-out (LOPO) evaluation the system reports 0.978 contact-level positive predictive value (PPV), 100% region-level accuracy, and a 15.1% improvement over the seizure-onset-zone-as-EZ baseline (p < 0.05); an external dataset yields 0.857 contact-level PPV. The approach requires only one night of interictal sleep data.

Significance. If the reported performance is reproducible, the work would represent a clinically meaningful advance: current seizure-freedom rates after resective surgery remain below 50%, and a method that reliably localizes the EZ from limited interictal data could shorten invasive monitoring and improve outcomes. The scale of the training corpus, the external validation, and the multimodal integration via a language model are notable strengths.

major comments (3)
  1. [Abstract / Methods] Abstract and Methods (training strategy paragraph): the central claim that anchoring biomarker extraction on non-epileptic signals yields a reliable, generalizable EZ marker rests on an untested assumption; the manuscript must demonstrate that this anchoring does not simply recover non-specific signal features and must report ablation results that isolate its contribution to the 0.978 PPV.
  2. [Results] Results (LOPO and external validation sections): the reported p < 0.05 for the 15.1% improvement and the 0.978 / 0.857 PPV figures lack accompanying details on the exact statistical test, correction for multiple comparisons across contacts/regions/patients, and patient-level characteristics of the external cohort; these omissions are load-bearing for the superiority claim.
  3. [Methods] Methods (model architecture and hyperparameters): no description is supplied of the signal foundation model architecture, pre-training objective, hyperparameter choices, or data exclusion criteria, preventing assessment of whether the high LOPO performance is reproducible or over-fit to the MNI cohort.
minor comments (2)
  1. [Abstract] Abstract: the acronym SOZ is introduced without expansion.
  2. [Abstract] Abstract: the external dataset size, number of patients, and their clinical characteristics should be stated to allow evaluation of the 0.857 PPV result.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below and will revise the manuscript accordingly to improve clarity, reproducibility, and evidential support for the central claims.

read point-by-point responses

  1. Referee: [Abstract / Methods] Abstract and Methods (training strategy paragraph): the central claim that anchoring biomarker extraction on non-epileptic signals yields a reliable, generalizable EZ marker rests on an untested assumption; the manuscript must demonstrate that this anchoring does not simply recover non-specific signal features and must report ablation results that isolate its contribution to the 0.978 PPV.

    Authors: We agree that explicit ablation experiments are required to isolate the contribution of anchoring on non-epileptic signals and to rule out recovery of non-specific features. The current manuscript describes the training strategy but does not contain such ablations. In the revision we will add ablation results under the same LOPO protocol, directly comparing the anchored model against a non-anchored counterpart and quantifying the resulting change in contact-level PPV. revision: yes

  2. Referee: [Results] Results (LOPO and external validation sections): the reported p < 0.05 for the 15.1% improvement and the 0.978 / 0.857 PPV figures lack accompanying details on the exact statistical test, correction for multiple comparisons across contacts/regions/patients, and patient-level characteristics of the external cohort; these omissions are load-bearing for the superiority claim.

    Authors: We will expand the Results section to supply the missing statistical details, including the precise test used for the reported p-value, any correction procedure applied across contacts or regions, and a supplementary table of patient-level characteristics for the external cohort (age, epilepsy type, recording duration, number of contacts). revision: yes

  3. Referee: [Methods] Methods (model architecture and hyperparameters): no description is supplied of the signal foundation model architecture, pre-training objective, hyperparameter choices, or data exclusion criteria, preventing assessment of whether the high LOPO performance is reproducible or over-fit to the MNI cohort.

    Authors: The initial submission omitted these specifications. The revised Methods section will provide a full description of the signal foundation model architecture, pre-training objective, hyperparameter choices, and data exclusion criteria so that readers can evaluate reproducibility and potential cohort-specific overfitting. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper reports training a signal foundation model on a large corpus of sEEG recordings (including mixed outcomes) with biomarker extraction anchored to non-epileptic signals, followed by a language model for multimodal fusion. Performance is measured via leave-one-patient-out evaluation on held-out patients plus an external dataset, yielding contact-level PPV figures that are not algebraically or definitionally identical to any fitted parameter or input subset. No equations, self-citations, or ansatzes are invoked that would force the reported metrics by construction; the evaluation protocol uses independent test partitions, rendering the derivation self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; therefore no specific free parameters, axioms, or invented entities can be extracted. The central claim rests on the unverified assumption that the described training procedure yields a generalizable biomarker.

pith-pipeline@v0.9.1-grok · 5763 in / 1226 out tokens · 37810 ms · 2026-06-26T10:29:03.707444+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

60 extracted references · 3 linked inside Pith

  1. [1]

    Epileptic Disorders23(1), 17–39 (2021)

    Surges, R., Shmuely, S., Dietze, C., Ryvlin, P., Thijs, R.D.: Identifying patients with epilepsy at high risk of cardiac death: signs, risk factors and initial management of high risk of cardiac death. Epileptic Disorders23(1), 17–39 (2021)

    2021

  2. [2]

    Nature Medicine30(5), 1292–1299 (2024)

    Jamiolkowski, R.M., Nguyen, Q.-A., Farrell, J.S., McGinn, R.J., Hartmann, D.A., Nirschl, J.J., Sanchez, M.I., Buch, V.P., Soltesz, I.: The fasciola cinereum of the hippocampal tail as an interventional target in epilepsy. Nature Medicine30(5), 1292–1299 (2024)

    2024

  3. [3]

    Journal of experimental pharmacology, 265–290 (2021)

    Rahim, F., Azizimalamiri, R., Sayyah, M., Malayeri, A.: Experimental therapeutic strategies in epilepsies using anti-seizure medications. Journal of experimental pharmacology, 265–290 (2021)

    2021

  4. [4]

    Nature Reviews Neurology8(12), 669–677 (2012)

    Wiebe, S., Jette, N.: Pharmacoresistance and the role of surgery in difficult to treat epilepsy. Nature Reviews Neurology8(12), 669–677 (2012)

    2012

  5. [5]

    Neurology43(8), 1612– 1612 (1993)

    Engel Jr, J.: Update on surgical treatment of the epilepsies: summary of the second international palm desert conference on the surgical treatment of the epilepsies (1992). Neurology43(8), 1612– 1612 (1993)

    1992

  6. [6]

    Epilepsia42(2), 282–286 (2001)

    Wieser, H.G., Blume, W.T., Fish, D., Goldensohn, E., Hufnagel, A., Kahane, P., L¨ uders, H., Pedley, T.A., Sutherling, W.: Proposal for a new classification of outcome with respect to epileptic seizures following epilepsy surgery. Epilepsia42(2), 282–286 (2001)

    2001

  7. [7]

    Journal of Neurology, Neurosurgery & Psychiatry (2026)

    Avigdor, T., Ho, A., Moye, M., Davalan, W., Minato, E., Hannan, S., Holden, T., Bouchet, T., Wang, Y.L., Jaber, K., Khweileh, M., Kaplan, S., Travnicek, V., Carlson, D., Frauscher, B.: Epilepsy surgery outcomes and their determinants: a systematic review and individual patient data meta-analysis. Journal of Neurology, Neurosurgery & Psychiatry (2026)

    2026

  8. [8]

    Seizure44, 217–224 (2017)

    Malmgren, K., Edelvik, A.: Long-term outcomes of surgical treatment for epilepsy in adults with regard to seizures, antiepileptic drug treatment and employment. Seizure44, 217–224 (2017)

    2017

  9. [9]

    Nature Communications15(1), 5253 (2024)

    Jaber, K., Avigdor, T., Mansilla, D., Ho, A., Thomas, J., Abdallah, C., Chabardes, S., Hall, J., Minotti, L., Kahane, P.,et al.: A spatial perturbation framework to validate implantation of the epileptogenic zone. Nature Communications15(1), 5253 (2024)

    2024

  10. [10]

    Clinical Neurophysiology134, 88–99 (2022)

    Klimes, P., Peter-Derex, L., Hall, J., Dubeau, F., Frauscher, B.: Spatio-temporal spike dynamics predict surgical outcome in adult focal epilepsy. Clinical Neurophysiology134, 88–99 (2022)

    2022

  11. [11]

    Molecular Imaging & Biology4(5), 338–351 (2002)

    Casse, R., Rowe, C.C., Newton, M., Berlangieri, S.U., Scott, A.M.: Positron emission tomography and epilepsy. Molecular Imaging & Biology4(5), 338–351 (2002)

    2002

  12. [12]

    Seizure41, 196–200 (2016)

    Ramantani, G., Maillard, L., Koessler, L.: Correlation of invasive eeg and scalp eeg. Seizure41, 196–200 (2016)

    2016

  13. [13]

    OpenNeuro (2026)

    Zhang, Y., Daida, A., Liu, L., Kuroda, N., Ding, Y., Oana, S., Monsoor, T., Duan, C., Hussain, S.A., Qiao, J.X., Salamon, N., Fallah, A., Sim, M.S., Sankar, R., Staba, R.J., Engel, J.J., Asano, E., Roychowdhury, V., Nariai, H.: Open iEEG Dataset (Pediatric iEEG, Wayne State University and UCLA). OpenNeuro (2026)

    2026

  14. [14]

    OpenNeuro (2025)

    Hatano, K., Kuroda, N., Uda, H., Sakakura, K., Cools, M.J., Luat, A.F., Osawa, S.-I., Nemoto, H., Ukishiro, K., Endo, H., Nakasato, N., Takayama, Y., Iijima, K., Iwasaki, M., Asano, E.: iEEG Comprehensive HFA Model Part 1. OpenNeuro (2025)

    2025

  15. [15]

    Epilepsy Currents18(1), 12–16 (2018)

    Jehi, L.: The epileptogenic zone: concept and definition. Epilepsy Currents18(1), 12–16 (2018)

    2018

  16. [16]

    Epileptic Disorders8, 1–9 (2006)

    L¨ uders, H.O., Najm, I., Nair, D., Widdess-Walsh, P., Bingman, W.: The epileptogenic zone: general principles. Epileptic Disorders8, 1–9 (2006)

    2006

  17. [17]

    Epilepsia57(3), 386–401 (2016)

    Mullin, J.P., Shriver, M., Alomar, S., Najm, I., Bulacio, J., Chauvel, P., Gonzalez-Martinez, J.: Is seeg safe? a systematic review and meta-analysis of stereo-electroencephalography–related complications. Epilepsia57(3), 386–401 (2016)

    2016

  18. [18]

    Epilepsia60(12), 2404–2415 (2019)

    Klimes, P., Cimbalnik, J., Brazdil, M., Hall, J., Dubeau, F., Gotman, J., Frauscher, B.: Nrem sleep is the state of vigilance that best identifies the epileptogenic zone in the interictal electroencephalo- gram. Epilepsia60(12), 2404–2415 (2019)

    2019

  19. [19]

    Journal of Neural Engineering (2024)

    Balaji, S.S., Parhi, K.K.: Seizure onset zone (soz) identification using effective brain connectivity of epileptogenic networks. Journal of Neural Engineering (2024)

    2024

  20. [20]

    Brain Communications7(3), 140 (2025)

    Nejedly, P., Hrtonova, V., Pail, M., Cimbalnik, J., Daniel, P., Travnicek, V., Dolezalova, I., Mivalt, F., Kremen, V., Jurak, P., Worrell, G.A., Frauscher, B., Klimes, P., Brazdil, M.: Leveraging interictal multimodal features and graph neural networks for automated planning of epilepsy surgery. Brain Communications7(3), 140 (2025)

    2025

  21. [21]

    Epilepsia65(10), 2935–2945 (2024)

    Klimes, P., Nejedly, P., Hrtonova, V., Cimbalnik, J., Travnicek, V., Pail, M., Peter-Derex, L., Hall, J., Pana, R., Halamek, J.,et al.: Interictal stereo-electroencephalography features below 45 Hz are sufficient for correct localization of the epileptogenic zone and postsurgical outcome prediction. Epilepsia65(10), 2935–2945 (2024)

    2024

  22. [22]

    Makaram, N., Gupta, S., Pesce, M., Bolton, J., Stone, S., Haehn, D., Pomplun, M., Papadelis, C., Pearl, P., Rotenberg, A.,et al.: Deep learning-based visual complexity analysis of electroencephalog- raphy time-frequency images: Can it localize the epileptogenic zone in the brain? Algorithms16(12), 567 (2023)

    2023

  23. [23]

    Neurology107(1), 218154 (2026)

    Abdallah, C., Cai, Z., Rammal, S., Aron, O., Ellenrieder, N., Chen, G., Hannan, S., Thomas, J., Kahane, P., Minotti, L.,et al.: Quantifying the effect of seizure-onset zone resection on epilepsy surgery outcome: A bayesian causal analysis. Neurology107(1), 218154 (2026)

    2026

  24. [24]

    In: International Conference on Learning Representations (2021)

    Tang, S., Dunnmon, J., Saab, K.K., Zhang, X., Huang, Q., Dubost, F., Rubin, D., Lee-Messer, C.: Self-supervised graph neural networks for improved electroencephalographic seizure analysis. In: International Conference on Learning Representations (2021)

    2021

  25. [25]

    In: Proceedings of the AAAI Conference on Artificial Intelligence, vol

    Ho, T.K.K., Armanfard, N.: Self-supervised learning for anomalous channel detection in eeg graphs: Application to seizure analysis. In: Proceedings of the AAAI Conference on Artificial Intelligence, vol. 37, pp. 7866–7874 (2023)

    2023

  26. [26]

    In: Proceedings of the Forty-first Conference on Uncertainty in Artificial Intelligence

    Ho, T.K.K., Armanfard, N.: Contaminated multivariate time-series anomaly detection with spatio- temporal graph conditional diffusion models. In: Proceedings of the Forty-first Conference on Uncertainty in Artificial Intelligence. Proceedings of Machine Learning Research, vol. 286, pp. 1710–1729. PMLR, Rio de Janeiro, Brazil (2025)

    2025

  27. [27]

    In: European Conference on Artificial Intelligence (2024)

    Lai, T., Ho, T.K.K., Armanfard, N.: Open-set multivariate time-series anomaly detection. In: European Conference on Artificial Intelligence (2024)

    2024

  28. [28]

    Neuron111(23), 3710–3715 (2023)

    Rahimzadeh, V., Jones, K.M., Majumder, M.A., Kahana, M.J., Rutishauser, U., Williams, Z.M., Cash, S.S., Paulk, A.C., Zheng, J., Beauchamp, M.S.,et al.: Benefits of sharing neurophysiology data from the BRAIN initiative research opportunities in humans consortium. Neuron111(23), 3710–3715 (2023)

    2023

  29. [29]

    In: Proceedings of NAACL-HLT, vol

    Devlin, J., Chang, M.-W., Lee, K., Toutanova, K.: Bert: Pre-training of deep bidirectional transformers for language understanding. In: Proceedings of NAACL-HLT, vol. 1 (2019)

    2019

  30. [30]

    In: Oh, A., Naumann, T., Globerson, A., Saenko, K., Hardt, M., Levine, S

    Yang, C., Westover, M., Sun, J.: Biot: Biosignal transformer for cross-data learning in the wild. In: Oh, A., Naumann, T., Globerson, A., Saenko, K., Hardt, M., Levine, S. (eds.) Advances in Neural Information Processing Systems, vol. 36, pp. 78240–78260 (2023)

    2023

  31. [31]

    arXiv preprint arXiv:2507.05201 (2025)

    Sellergren, A., Kazemzadeh, S., Jaroensri, T., Kiraly, A., Traverse, M., Kohlberger, T., Xu, S., Jamil, F., Hughes, C., Lau, C., et al.: MedGemma technical report. arXiv preprint arXiv:2507.05201 (2025)

    Pith/arXiv arXiv 2025

  32. [32]

    Nature Medicine31(3), 943–950 (2025)

    Singhal, K., Tu, T., Gottweis, J., Sayres, R., Wulczyn, E., Amin, M., Hou, L., Clark, K., Pfohl, S.R., Cole-Lewis, H.,et al.: Toward expert-level medical question answering with large language models. Nature Medicine31(3), 943–950 (2025)

    2025

  33. [33]

    arXiv preprint arXiv:2305.15525 (2023)

    Liu, X., McDuff, D., Kovacs, G., Galatzer-Levy, I., Sunshine, J., Zhan, J., Poh, M.-Z., Liao, S., Di Achille, P., Patel, S.: Large language models are few-shot health learners. arXiv preprint arXiv:2305.15525 (2023)

    arXiv 2023

  34. [34]

    Advances in Neural Information Processing Systems35, 24824–24837 (2022)

    Wei, J., Wang, X., Schuurmans, D., Bosma, M., Xia, F., Chi, E., Le, Q.V., Zhou, D.,et al.: Chain- of-thought prompting elicits reasoning in large language models. Advances in Neural Information Processing Systems35, 24824–24837 (2022)

    2022

  35. [35]

    Clinical Neurophysiology169, 33–46 (2025)

    Hrtonova, V., Nejedly, P., Travnicek, V., Cimbalnik, J., Matouskova, B., Pail, M., Peter-Derex, L., Grova, C., Gotman, J., Halamek, J.,et al.: Metrics for evaluation of automatic epileptogenic zone localization in intracranial electrophysiology. Clinical Neurophysiology169, 33–46 (2025)

    2025

  36. [36]

    Ellenrieder, N., Khoo, H.M., Dubeau, F., Gotman, J.: What do intracerebral electrodes measure? Clinical Neurophysiology132(5), 1105–1115 (2021)

    2021

  37. [37]

    Brain Communications5(6), 304 (2023)

    Dessert, G.E., Thio, B.J., Grill, W.M.: Optimization of patient-specific stereo-eeg recording sensitivity. Brain Communications5(6), 304 (2023)

    2023

  38. [38]

    Frontiers in Neurology16, 1685431 (2025)

    Reinacher, P.C., Altenm¨ uller, D.-M., Nakagawa, J.M., Li Hegner, Y., Antal, C.D., D¨ umpelmann, M., Demerath, T., Staack, A.M., Huppertz, H.-J., Doostkam, S.,et al.: Combined mri morphometry and source imaging guide placement of stereo-eeg electrodes in focal epilepsy with subtle or absent lesions. Frontiers in Neurology16, 1685431 (2025)

    2025

  39. [39]

    arXiv preprint arXiv:2106.11170 (2021)

    Song, Y., Jia, X., Yang, L., Xie, L.: Transformer-based spatial-temporal feature learning for eeg decoding. arXiv preprint arXiv:2106.11170 (2021)

    arXiv 2021

  40. [40]

    Neurology100(17), 1750–1762 (2023)

    Jing, J., Ge, W., Hong, S., Fernandes, M.B., Lin, Z., Yang, C., An, S., Struck, A.F., Herlopian, A., Karakis, I.,et al.: Development of expert-level classification of seizures and rhythmic and periodic patterns during eeg interpretation. Neurology100(17), 1750–1762 (2023)

    2023

  41. [41]

    In: 2022 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), pp

    Peh, W.Y., Yao, Y., Dauwels, J.: Transformer convolutional neural networks for automated artifact detection in scalp eeg. In: 2022 44th Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), pp. 3599–3602 (2022). IEEE

    2022

  42. [42]

    arXiv preprint arXiv:2601.03267 (2025)

    Singh, A., Fry, A., Perelman, A., Tart, A., Ganesh, A., El-Kishky, A., McLaughlin, A., Low, A., Ostrow, A., Ananthram, A., et al.: OpenAI GPT-5 system card. arXiv preprint arXiv:2601.03267 (2025)

    Pith/arXiv arXiv 2025

  43. [43]

    arXiv preprint arXiv:2410.21276 (2024)

    Hurst, A., Lerer, A., Goucher, A.P., Perelman, A., Ramesh, A., Clark, A., Ostrow, A., Welihinda, A., Hayes, A., Radford, A., et al.: GPT-4o system card. arXiv preprint arXiv:2410.21276 (2024)

    Pith/arXiv arXiv 2024

  44. [44]

    npj Digital Medicine8(1), 141 (2025)

    Xie, Q., Chen, Q., Chen, A., Peng, C., Hu, Y., Lin, F., Peng, X., Huang, J., Zhang, J., Keloth, V., et al.: Medical foundation large language models for comprehensive text analysis and beyond. npj Digital Medicine8(1), 141 (2025)

    2025

  45. [45]

    National Academies Press, Washington, DC (2006)

    Altevogt, B.M., Colten, H.R.: Sleep Disorders and Sleep Deprivation: An Unmet Public Health Problem. National Academies Press, Washington, DC (2006)

    2006

  46. [46]

    Epilepsia38(12), 1300–1314 (1997)

    Spanedda, F., Cendes, F., Gotman, J.: Relations between eeg seizure morphology, interhemispheric spread, and mesial temporal atrophy in bitemporal epilepsy. Epilepsia38(12), 1300–1314 (1997)

    1997

  47. [47]

    NPJ Digital Medicine7(1), 20 (2024)

    Savage, T., Nayak, A., Gallo, R., Rangan, E., Chen, J.H.: Diagnostic reasoning prompts reveal the potential for large language model interpretability in medicine. NPJ Digital Medicine7(1), 20 (2024)

    2024

  48. [48]

    Brain141(4), 1130–1144 (2018)

    Frauscher, B., Von Ellenrieder, N., Zelmann, R., Dolezalov´ a, I., Minotti, L., Olivier, A., Hall, J., Hoff- mann, D., Nguyen, D.K., Kahane, P.,et al.: Atlas of the normal intracranial electroencephalogram: neurophysiological awake activity in different cortical areas. Brain141(4), 1130–1144 (2018)

    2018

  49. [49]

    Journal of Clinical Sleep Medicine 8(5), 597–619 (2012)

    Berry, R.B., Budhiraja, R., Gottlieb, D.J., Gozal, D., Iber, C., Kapur, V.K., Marcus, C.L., Mehra, R., Parthasarathy, S., Quan, S.F.,et al.: Rules for scoring respiratory events in sleep: update of the 2007 aasm manual for the scoring of sleep and associated events: deliberations of the sleep apnea definitions task force of the american academy of sleep m...

    2007

  50. [50]

    In: Joint European Conference on Machine Learning and Knowledge Discovery in Databases, pp

    Hojjati, H., Sadeghi, M., Armanfard, N.: Multivariate time-series anomaly detection with temporal self-supervision and graphs: Application to vehicle failure prediction. In: Joint European Conference on Machine Learning and Knowledge Discovery in Databases, pp. 242–259 (2023). Springer

    2023

  51. [51]

    IEEE Transactions on Pattern Analysis and Machine Intelligence (2025)

    Ho, T.K.K., Karami, A., Armanfard, N.: Graph anomaly detection in time series: A survey. IEEE Transactions on Pattern Analysis and Machine Intelligence (2025). Preprint at arXiv:2302.00058

    arXiv 2025

  52. [52]

    In: Advances in Neural Information Processing Systems, vol

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L., Polosukhin, I.: Attention is all you need. In: Advances in Neural Information Processing Systems, vol. 30 (2017)

    2017

  53. [53]

    In: International Conference on Machine Learning, pp

    Katharopoulos, A., Vyas, A., Pappas, N., Fleuret, F.: Transformers are rnns: Fast autoregressive transformers with linear attention. In: International Conference on Machine Learning, pp. 5156–5165 (2020). PMLR

    2020

  54. [54]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    He, K., Fan, H., Wu, Y., Xie, S., Girshick, R.: Momentum contrast for unsupervised visual repre- sentation learning. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 9729–9738 (2020)

    2020

  55. [55]

    In: The Eleventh International Conference on Learning Representations (2023)

    Pan, J., Zhou, P., Yan, S.: Towards understanding why mask reconstruction pretraining helps in downstream tasks. In: The Eleventh International Conference on Learning Representations (2023)

    2023

  56. [56]

    Seizure22(7), 493–501 (2013)

    Surges, R., Elger, C.E.: Reoperation after failed resective epilepsy surgery. Seizure22(7), 493–501 (2013)

    2013

  57. [57]

    Neurosurgery75(6), 648–656 (2014)

    Englot, D.J., Raygor, K.P., Molinaro, A.M., Garcia, P.A., Knowlton, R.C., Auguste, K.I., Chang, E.F.: Factors associated with failed focal neocortical epilepsy surgery. Neurosurgery75(6), 648–656 (2014)

    2014

  58. [58]

    JAMA Neurology76(4), 462–469 (2019)

    Andrews, J.P., Gummadavelli, A., Farooque, P., Bonito, J., Arencibia, C., Blumenfeld, H., Spencer, D.D.: Association of seizure spread with surgical failure in epilepsy. JAMA Neurology76(4), 462–469 (2019)

    2019

  59. [59]

    In: International Conference on Learning Representations (2020)

    Ruff, L., Vandermeulen, R.A., G¨ ornitz, N., Binder, A., M¨ uller, E., M¨ uller, K.-R., Kloft, M.: Deep semi-supervised anomaly detection. In: International Conference on Learning Representations (2020)

    2020

  60. [60]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp

    Cui, Y., Jia, M., Lin, T.-Y., Song, Y., Belongie, S.: Class-balanced loss based on effective num- ber of samples. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 9268–9277 (2019)

    2019