Recent advancements in data collection through seizure diaries, invasive and non-invasive devices, and implementing computational algorithms for identifying seizure patterns and likelihood such as statistical and machine learning algorithms have made reliable seizure forecasting increasingly feasible (Meisel et al., ; Brinkmann et al., ; Nasseri et al., ; Stirling et al., ). This holds great promise for improved seizure management. However, several significant challenges remain such as high false alarm rates that undermine the reliability of current forecasting algorithms (Karoly et al., ), and the variability in seizure patterns among individuals, which makes it difficult to create a generalized model. Additionally, the quality of recorded physiological data is crucial (Nasseri et al., ; Böttcher et al., ); while non-invasive devices can provide valuable data, they are not always acceptable to all patients. Integrating various types of data, such as physiological signals, sleep patterns, and mood information, into a single forecasting model requires substantial data processing power and sophisticated algorithms. Moreover, implementing seizure forecasting in real-life conditions necessitates evaluating the acceptability of algorithm performance metrics among patients and clinicians (Grzeskowiak and Dumanis, ). These challenges underscore the need for ongoing research and innovation in the field of seizure forecasting.
One of the primary challenges with current seizure forecasting algorithms is their high false alarm rate (FAR), which limits clinical applicability and acceptability to patients (Alotaiby et al., ; Beniczky et al., ). High FAR is a critical barrier to the widespread clinical adoption and utility of seizure forecasting technologies necessitating development of more robust and accurate prediction models. In the study Segal et al. researchers applied a risk-controlling prediction (RCP) calibration method called Learn then Test (LTT) to address this issue. The calibration algorithm was first validated with synthetic data and then tested on scalp EEG recordings. A convolutional neural network was used to assess seizure risk by defining the preictal state as the period from 60 min to 30 s before seizure onset. By implementing LTT as a post-processing step on the test dataset, an average reduction of 92% in the FAR was achieved. Although this calibration method improved model performance for some recordings, the results were still not sufficient for clinical relevance.
On the other hand, to develop reliable forecasting algorithms, identifying biomarkers and features that capture the non-linear and non-stationary nature of physiological signals is another promising area of investigation (Brinkmann et al., ; Meisel et al., ; Nasseri et al., ). A 2014 paper (Wang and Lyu, ) derived comprehensive feature vectors from EEG signals for efficient prediction and examined the relevance of primary amplitude and frequency components to a patient's seizure occurrence. The study by Chen et al., highlights the significance of time-frequency features extracted from cEEG in forecasting seizures. Time-frequency and power spectrum analysis were applied to investigate periodic discharges (PDs) patterns and their relation to seizure prediction. The study demonstrated that high spectral power predicted a high risk of seizures, while low spectral power was associated with a lower risk. However, a prospective study that includes a larger cohort of patients is needed to confirm these findings.
During the first international seizure prediction workshop, the seizure prediction community established standards for datasets to be used in the development and evaluation of seizure prediction algorithms (Lehnertz and Litt, ). As a result, several EEG datasets adhering to these guidelines were created and are recommended for research groups working in the area of seizure prediction (Wagenaar et al., ). Further datasets were made available by the community (Brinkmann et al., ; Kuhlmann et al., ). In the current Research Topic, Andrade et al. compared the performance of a patient-specific seizure prediction algorithm across four open access databases in a standardized way. The study distinguishes between sample-based approaches disregarding the temporal aspect of seizures, and alarm-based approaches, which aim to simulate real-life conditions. In this more realistic scenario results were less promising and shows the importance of rigorous testing conditions. However, the existing databases also have deficits, as they contain EEG recordings of only a few seizures for many patients. Datasets with ultra-long term recordings with subcutaneous (Pal Attia et al., ; Viana et al., ) or intracranial EEG (Kuhlmann et al., ) offer the possibility to overcome some of these limitations, although these continuous ultra-long-term datasets remain rare due to the challenges in acquiring such data. Efforts are underway to overcome these barriers (Khambhati et al., ).
A lower burden for patients is achieved by using wearable devices or intelligent (sensor-integrated) clothing during daily routines. Gagliano et al. used a smart shirt with a single-lead ECG, two respiratory bands and a detachable telemetry device with a three-axis accelerometer to investigate the relation between sleep efficiency and the occurrence of epileptogenic seizures. Relationships between vigilance states, sleep stages and epilepsy are known for many years and circadian profiles of seizure occurrence are not only results of recent research (Khan et al., ) but are still not completely understood and remain an active Research Topic, particularly in relation to ictal and interictal EEG activity (Peter-Derex et al., ). In this study, four sleep metrics were derived from the sensor data of the smart shirt: total sleep duration, sleep latency, time of waking after sleep onset, and sleep efficiency. It could be shown, that sleep quality, especially sleep efficiency is lower in nights before a seizure. These results complement those obtained using a wristband wearable (Stirling et al., ), showing that sleep onset and offset times were significantly associated with heightened seizure risk the following day for more participants than changes in sleep duration the night before. Overall these studies emphasize the richness of information, which can be derived from mobile recording devices contributing to seizure forecasting in real life settings.
Seizure forecasting holds the promise of significantly reducing the burden of epilepsy and enhancing the autonomy and quality of life for those affected. To fully realize this potential, further improvements in accuracy, reduction of patient burden, and integration with targeted interventions are essential.
Statements
Author contributions
MN: Writing – original draft, Conceptualization, Investigation, Writing – review & editing. CG: Writing – original draft, Writing – review & editing. BB: Writing – original draft, Writing – review & editing. MD: Conceptualization, Supervision, Writing – original draft, Writing – review & editing.
Funding
The author(s) declare that no financial support was received for the research, authorship, and/or publication of this article.
Conflict of interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The author(s) declared that they were an editorial board member of Frontiers, at the time of submission. This had no impact on the peer review process and the final decision.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
References
1
AlotaibyT. N.AlshebeiliS. A.AlshawiT.AhmadI.Abd El-SamieF. E. (2014). EEG seizure detection and prediction algorithms: a survey. EURASIP J. Adv. Sign. Process. 2014, 1–21. 10.1186/1687-6180-2014-183
2
BeniczkyS.WiebeS.JeppesenJ.TatumW. O.BrazdilM.WangY.et al. (2021). Automated seizure detection using wearable devices: a clinical practice guideline of the International League Against Epilepsy and the International Federation of Clinical Neurophysiology. Clin. Neurophysiol. 132, 1173–1184. 10.1016/j.clinph.2020.12.009
3
BöttcherS.VielufS.BrunoE.JosephB.EpitashviliN.BiondiA.et al. (2022). Data quality evaluation in wearable monitoring. Sci. Rep. 12:21412. 10.1038/s41598-022-25949-x
4
BrinkmannB. H.KarolyP. J.NurseE. S.DumanisS. B.NasseriM.VianaP. F.et al. (2021). Seizure diaries and forecasting with wearables: epilepsy monitoring outside the clinic. Front. Neurol. 12:690404. 10.3389/fneur.2021.690404
5
BrinkmannB. H.WagenaarJ.AbbotD.AdkinsP.BosshardS. C.ChenM.et al. (2016). Crowdsourcing reproducible seizure forecasting in human and canine epilepsy. Brain139, 1713–1722. 10.1093/brain/aww045
6
GrzeskowiakC. L.DumanisS. B. (2021). Seizure forecasting: patient and caregiver perspectives. Front. Neurol. 12:717428. 10.3389/fneur.2021.717428
7
KarolyP. J.CookM. J.MaturanaM.NurseE. S.PayneD.BrinkmannB. H.et al. (2020). Forecasting cycles of seizure likelihood. Epilepsia61, 776–786. 10.1111/epi.16485
8
KhambhatiA. N.ChangE. F.BaudM. O.RaoV. R. (2024). Hippocampal network activity forecasts epileptic seizures. Nat. Med.6, 1–4. 10.1038/s41591-024-03149-6
9
KhanS.NobiliL.KhatamiR.LoddenkemperT.CajochenC.DijkD.-J.et al. (2018). Circadian rhythm and epilepsy. Lancet Neurol. 17, 1098–1108. 10.1016/S1474-4422(18)30335-1
10
KuhlmannL.KarolyP.FreestoneD. R.BrinkmannB. H.TemkoA.BarachantA.et al. (2018). Epilepsyecosystem.org: crowd-sourcing reproducible seizure prediction with long-term human intracranial EEG. Brain141, 2619–2630. 10.1093/brain/awy210
11
LehnertzK.LittB. (2005). The first international collaborative workshop on seizure prediction: summary and data description. Clin. Neurophysiol. 116, 493–505. 10.1016/j.clinph.2004.08.020
12
MeiselC.El AtracheR.JacksonM.SchubachS.UfongeneC.LoddenkemperT. (2020). Machine learning from wristband sensor data for wearable, noninvasive seizure forecasting. Epilepsia61, 2653–2666. 10.1111/epi.16719
13
NasseriM.NurseE.GlasstetterM.BöttcherS.GreggN. M.Laks NandakumarA.et al. (2020). Signal quality and patient experience with wearable devices for epilepsy management. Epilepsia61, S25–S35. 10.1111/epi.16527
14
NasseriM.Pal AttiaT.JosephB.GreggN. M.NurseE. S.VianaP. F.et al. (2021). Ambulatory seizure forecasting with a wrist-worn device using long-short term memory deep learning. Sci. Rep. 11:21935. 10.1038/s41598-021-01449-2
15
Pal AttiaT.VianaP. F.NasseriM.Duun-HenriksenJ.BiondiA.WinstonJ. S.et al. (2023). Seizure forecasting using minimally invasive, ultra-long-term subcutaneous EEG: generalizable cross-patient models. Epilepsia64, S114–S123. 10.1111/epi.17265
16
Peter-DerexL.KlimesP.LatreilleV.BouhadounS.DubeauF.FrauscherB. (2020). Sleep disruption in epilepsy: ictal and interictal epileptic activity matter. Ann. Neurol. 88, 907–920. 10.1002/ana.25884
17
StirlingR. E.GraydenD. B.D'SouzaW.CookM. J.NurseE.FreestoneD. R.et al. (2021). Forecasting seizure likelihood with wearable technology. Front. Neurol. 12:704060. 10.3389/fneur.2021.704060
18
StirlingR. E.HidajatC. M.GraydenD. B.D'SouzaW. J.Naim-FeilJ.DellK. L.et al. (2023). Sleep and seizure risk in epilepsy: bed and wake times are more important than sleep duration. Brain146, 2803–2813. 10.1093/brain/awac476
19
VianaP. F.Pal AttiaT.NasseriM.Duun-HenriksenJ.BiondiA.WinstonJ. S.et al. (2023). Seizure forecasting using minimally invasive, ultra-long-term subcutaneous electroencephalography: individualized intrapatient models. Epilepsia64, S124–S133. 10.1111/epi.17252
20
WagenaarJ. B.WorrellG. A.IvesZ.MatthiasD.LittB.Schulze-BonhageA. (2015). Collaborating and sharing data in epilepsy research. J. Clin. Neurophysiol. 32, 235–239. 10.1097/WNP.0000000000000159
21
WangN.LyuM. R. (2014). Extracting and selecting distinctive EEG features for efficient epileptic seizure prediction. IEEE J. Biomed. Health Informat. 19, 1648–1659. 10.1109/JBHI.2014.2358640
Summary
Keywords
seizure forecasting algorithms, seizure forecasting devices, seizure forecasting biomarkers, data collection, epilepsy
Citation
Nasseri M, Grzeskowiak C, Brinkmann BH and Dümpelmann M (2024) Editorial: Seizure forecasting tools, biomarkers and devices. Front. Neurosci. 18:1470640. doi: 10.3389/fnins.2024.1470640
Received
25 July 2024
Accepted
08 August 2024
Published
28 August 2024
Volume
18 - 2024
Edited and reviewed by
Guo-Yuan Yang, Shanghai Jiao Tong University, China
Updates
Copyright
© 2024 Nasseri, Grzeskowiak, Brinkmann and Dümpelmann.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Benjamin H. Brinkmann brinkmann.benjamin@mayo.edu
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.