71

Preoperative perivascular space asymmetry predicts memory trajectories after temporal lobe epilepsy surgery

Thaera Arafat, Joanna Bartkiewicz, Luisa Delazer, Nicholas Fearns, Kazuki Fukuma, Matthias Koepp, Adolfo Mazzeo, Alicija Mrzyglod, Isha Puntambekar, Michael Schubert, Maria Thom

Background: MRI-visible perivascular spaces (PVS) are markers of glymphatic dysfunction that are increased in temporal lobe epilepsy (TLE). We investigated whether preoperative hemispheric PVS asymmetry is associated with verbal and visual memory trajectories after TLE surgery.

Methods: We studied 141 patients from a TLE surgery registry (56% left TLE; 54% female; mean age 38 years) with one-year neuropsychological follow-up; 47 also completed long-term assessment (mean 9.3 years). Two laterality indices (LI) were computed as (ipsilateral − contralateral) / (ipsilateral + contralateral), where ipsilateral denotes the side of the epileptic focus and positive values indicate greater PVS burden at the seizure onset zone: temporal PVS volume (T-Vol LI) and non-temporal PVS volume (nT-Vol LI). Verbal memory (word list learning + delayed recall + story delayed recall) and visual memory (design learning + delayed recall) from the Adult BMIPB battery were summarised as PCA composites (PC1: 62–77% variance, α=0.51–0.80). Outcomes were z-score change scores (postoperative − preoperative); positive values indicate less decline. Standardised OLS regression adjusted for sex, age, surgery side, and domain-specific baseline. Tau burden was quantified by AT8 immunoreactivity in resection specimens (N=108). FDR correction (Benjamini–Hochberg) was applied across 8 tests.

Results: Higher temporal PVS asymmetry (T-Vol LI) was associated with less verbal memory decline one year after surgery (β=+0.27, p=0.043, FDR=0.172), while greater non-temporal PVS asymmetry (nT-Vol LI) was associated with less long-term visual memory decline (β=+0.51, p=0.006, FDR=0.050), representing the only association to survive correction for multiple comparisons. In both cases, greater ipsilateral PVS predominance was associated with better preservation of memory. The association between non-temporal PVS asymmetry and postoperative memory trajectory diminished with increasing age, with significant nT-Vol LI×age interactions for visual (p=0.040) and verbal memory (p=0.026), whereas temporal PVS asymmetry showed no significant age interaction (all p>0.25). Neither PVS laterality measure was associated with epilepsy severity (age at onset, disease duration, seizure frequency, number of antiseizure medications; all p>0.20) or AT8-positive tau pathology in resected tissue (all p>0.15).

Conclusions: Preoperative hemispheric PVS asymmetry was associated with postoperative memory trajectories independently of epilepsy severity and local tau pathology. Together with emerging evidence linking astrocytic dysfunction, impaired glutamate homeostasis and activity-dependent tau accumulation to cognitive decline, these findings suggest that MRI-visible PVS may capture an early disturbance of brain homeostasis that precedes overt neurodegeneration. Rather than reflecting accumulated pathology, hemispheric PVS asymmetry may identify differences in network resilience that influence long-term cognitive trajectories in TLE

72

A systematic review of brain network and connectivity measures that predict seizure outcome after epilepsy surgery

Damjan Veljanoski^1,2, Siddarth Kannan³, Aswin Chari^1,2, Martin Tisdall^1,2, Torsten Baldeweg^1,2, Rory Piper^1,2

¹Developmental Neurosciences, UCL Great Ormond Street Institute Of Child Health, ²Department of Neurosurgery, Great Ormond Street Institute of Child Health, ³NHS University Hospitals of Liverpool

Purpose: To identify and synthesise neuroimaging brain network and connectivity measures that predict postoperative seizure outcomes following epilepsy surgery, and to evaluate their methodological characteristics and predictive performance.

Methods: A PROSPERO-registered systematic review (CRD420251004246) was conducted in accordance with PRISMA. Embase, PubMed and Web of Science were searched without date or language restrictions. Two reviewers independently screened titles/abstracts and full texts. Eligible studies used neuroimaging network/connectivity analyses to predict seizure outcome after epilepsy surgery in adults or children. Data extracted included cohort and surgical characteristics, imaging modality and field strength, network construction/metrics, outcome definitions, and model evaluation (where applicable). Studies were classified as predictor-finding or predictor-modelling with risk of bias was assessed using QUIPS and PROBAST, respectively. Findings were narratively synthesised by connectivity domain, unit of analysis (global/subnetwork/nodal), anatomical grouping, and direction of association with outcome.

Results: Eighty-one studies were included; 86% were single-centre, 70% adult-only, and 69% focused on temporal lobe epilepsy. Median minimum follow-up was 12 months (range 3–60). The predominant intervention was resection (60%), followed by neuromodulation (RNS/DBS/VNS; 13%), combined LITT/resection (9%), and LITT-only (6%). Functional (35%), diffusion (25%), and metabolic (10%) analyses were most frequent. Most studies used 3T MRI (74%); none used 7T. Outcomes were variably defined (ILAE, Engel, responder rates), with median “good outcome” proportions ranging from 45% to 64% across outcome definitions. Among 18 predictor-modelling studies reporting accuracy, median outcome-prediction accuracy was 86%. Narrative synthesis indicated that favourable outcomes were generally associated with selective disconnection/reduction of pathological connectivity and preservation or reorganisation of key networks, whereas poor outcomes aligned with widespread/bilateral abnormalities and incomplete disruption of critical pathways.

Conclusions: Imaging-based connectivity and network metrics show promising accuracy for predicting postoperative seizure outcomes, but heterogeneity in cohorts, outcome definitions, and analytic pipelines limits comparability and generalisability.

73

Diffusion MRI abnormalities localise the epileptogenic zone in paediatric drug-resistant epilepsy

Damjan Veljanoski^1,2, Aswin Chari^1,2, Kiran Seunarine¹, Felice D’Arco³, Kshitij Mankad³, Zubair Tahir², Chris Clark¹, Martin Tisdall^1,2, Torsten Baldeweg^1,2, Rory Piper^1,2

¹Developmental Neurosciences, UCL Great Ormond Street Institute Of Child Health, ²Department of Neurosurgery, Great Ormond Street Hospital for Children NHS Trust, ³Department of Radiology, Great Ormond Street Hospital for Children NHS Trust

Background and objectives: Seizure freedom is achieved in 67% of children undergoing surgery for drug-resistant epilepsy, but reliable imaging biomarkers are needed to improve localization of the epileptogenic zone, patient selection and outcome prediction, particularly in MRI-negative patients. We evaluated a voxelwise diffusion MRI method for detecting microstructural abnormalities, testing whether abnormalities were associated with seizure freedom when resected, aligned with stereo-EEG-defined seizure onset and interictal zones, distinguished focal from distributed epileptic networks, including patients with and without an identified seizure onset zone, and localized the epileptogenic zone in MRI lesion-negative epilepsy.

Methods: We retrospectively studied children with epilepsy undergoing focal resection (n=110) or stereo-EEG implantation (n=61) between 2015-2023, and healthy controls aged 6–18 years (n=69), scanned using the same 3T MRI and diffusion protocol. Individual voxelwise z-score maps were generated for mean diffusivity (MD) and fractional anisotropy (FA) using age- and sex-adjusted normative models. Cluster-enhanced dominant abnormalities were identified. In the resection cohort, percentage overlap between abnormalities and resection masks was related to Engel outcome at ≥12 months. In the stereo-EEG cohort, diffusion z-scores were compared across seizure onset, interictal and remaining gray matter contacts. Group maps characterized clinical subgroups. In MRI lesion-negative patients, abnormalities were assessed for concordance with PET hypometabolism and stereo-EEG findings.

Results: Greater resection of the dominant MD abnormality predicted seizure freedom, with a stronger effect when restricted to patients with >0 overlap (mean differences 17% and 23%; Pperm=0.0086 and Pperm<0.001, respectively), and increased odds of Engel I outcome (OR 1.31 per 10%; AUC 0.723; Youden threshold 36%). FA abnormalities showed similar associations. In stereo-EEG patients, seizure onset contacts had higher MD and lower FA z-scores than remaining gray matter contacts (mean differences +0.65 and −0.18; Pperm<0.001 and Pperm=0.001), with interictal contacts intermediate. Abnormalities were more widespread and contralateral in patients without an identified seizure onset zone. In MRI lesion-negative patients, abnormalities co-localized with PET hypometabolism (20/28 fully, 6/28 partially) and were concordant with stereo-EEG findings (17/24 fully, 3/24 partially).

Discussion: Diffusion MRI identified abnormalities that aligned with invasive electrophysiology, associated with seizure freedom when resected, and co-localized with PET hypometabolism and stereo-EEG findings in lesion-negative epilepsy, supporting its potential to inform presurgical evaluation.

74

The Role of Low-field MRI in Epilepsy

Melody Asukile^1,2, Frederik Barkhof¹, Kelly Baruteau¹, Indran Davagnanam¹, Rui Yeow³, James Cole³, Josemir W Sander^1,3, John S Duncan¹

¹UCL Queen Square Institute of Neurology, ²Department of Internal Medicine, University of Zambia School of Medicine , ³UCL Hawkes Institute, ⁴Department of Neurology, West China Hospital, Sichuan University

Background: Magnetic Resonance Imaging (MRI) is essential in the diagnostic evaluation of epilepsy. Conventional MRI scanners are inaccessible in many low- and middle-income countries (LMIC). More affordable, portable low-field (LF) MRI scanners have the potential to improve access to brain imaging in resource-limited settings. Currently, image quality is limited but may be improved by Artificial intelligence (AI) algorithms. We explored the utility of LF-MRI in the diagnosis of causes of focal epilepsy.

Methods: Adults with focal epilepsy who had a 3T high-field (HF) MRI scan within 12 months underwent a 60-minute non-contrast LF-MRI on a Hyperfine® Swoop 64mT scanner. Scans were processed for image enhancement and segmentation with SuperSynth, an AI machine-learning tool. Three neuroradiologists independently read the native LF-MRI, AI-LF-MRI, and HF-MRI. Basic participant clinical information was provided.

Results: Of 39 participants, 51% were male (mean age 42 ± 12 years). 3T MRI diagnoses included: Non-lesional (10/39, 26%) Hippocampal sclerosis (HS) (8/39, 21%)

Post-surgical or gliotic (7/39, 18%), Long-term epilepsy associated tumours (LEAT) (7/39, 18%), Cavernoma (3/39, 8%), Cortical malformations (2/39, 6%), Hypothalamic hamartoma (1/39, 3%) and post-infectious calcification (1/39, 3%). There was moderate agreement among the neuroradiologists, with Cohen’s kappa = 0.33 (95% CI 0.18-0.48) and Gwet’s AC = 0.53 (95% CI 0.34-0.72). LF-MRI scan quality was reported as poor in 96%, with signal inhomogeneity the most frequent artefact (97%). LF-MRI was reported to be diagnostic in 80% of cases. Reporting confidence was 25% confident, 19% neutral, and 56% unconfident.

Concordance with 3T MRI was highest in LEAT, with complete agreement in 5/7 (71%) cases, and partial (2 of 3 raters) in 2/7 (29%), then gliotic lesions (4/7, 57%), and HS, complete agreement (3/8 cases, 37.5%) and partial (2/6, 37.5%), with overall agreement in HS ~75%. HS agreement was more likely in unequivocal cases. In non-lesional cases, agreement was in 2/10 and partial agreement in 5/10. Concordance with 3T reports was lowest for cavernomas (1/3) and malformations of cortical development (0/2). Reporting of the AI-LF-MRI is pending.

Conclusion: Qualitative analysis of LF-MRI scans by experienced neuroradiologists showed fair to moderate agreement. There were high rates of poor MRI quality and signal inhomogeneity. Gliotic lesions, tumours and HS showed the highest concordance with 3T-MRI. There was only partial agreement for other lesions and non-lesional cases. Analysis of AI-LF-MRI scans will help determine whether LF-MRI has a role in the evaluation of focal epilepsy in LMIC.

75

The global prevalence of malformations of cortical development in people with epilepsy: a systematic review and meta-analysis

Rebecca Maguire^1,2, Prem Jareonsettasin^1,2, Mina Hanifi^1,2, Professor Ley Sander^1,2

¹Queen’s Square Institute of Neurology, ²Social Determinants of Health in Epileptic Disorders (SdHIELD) Research Group

Introduction: Malformations of cortical development (MCDs) are important causes of drug-resistant epilepsy (DRE) worldwide. Understanding MCD epidemiology is essential to understanding risk factors, burden and the effects of potentially modifiable factors, including social determinants of health. We conducted a systematic review and meta-analysis to determine the prevalence of MCDs worldwide.

Methods: Nine databases (PubMed, Web of Science, Global Index Medicus, Embase, Scopus, Ovid Global Health, PsychINFO, LILACS, CNKI) were searched from 01/1980 to 09/2025 to identify cohorts of people with epilepsy who underwent neuroimaging. Inclusion criteria: original research articles from a single country, people with epilepsy who have undergone an MRI scan with a breakdown of radiological diagnoses. Studies restricted to a first seizure, acute symptomatic seizures or a selective differentiated cohort such as post-stroke epilepsy only, and with fewer than 50 people were excluded. A random-effects meta-analysis was performed to estimate the pooled prevalence of MCDs. Subgroup analysis examined the effects of drug-resistance status, age group, geographical region and national income status.

Results: Forty-one studies comprising 11,103 individuals across twenty-seven countries were included. 847 participants have MRI evidence of MCDs, giving a pooled prevalence of 7.88% (95% CI 6.10-10.12). There is substantial inter-study heterogeneity (I² = 91.6%). DRE cohorts have a higher prevalence (14.04%; 95% CI 10.23-18.97) compared to cohorts with mixed (6.41%; 95% CI 4.66-8.76) or unspecified (7.09%; 95% CI 4.78-10.38) drug-resistance status (p = 0.001). Children with DRE (11.41%; 95% CI 9.58-13.55) and adults with DRE (10.34%; 95% CI 7.76-13.65) have similar prevalence. Higher-income countries have a higher prevalence of MCDs in DRE (17.19%; 95% CI 10.56-26.73) than low- and middle-income countries (LMICs) (10.37%; 95% CI 8.12-13.15). Prevalence of MCDs in DRE for studies using 1.5 Tesla MRI is 11.72% (95% CI 9.95-14.31) and 3 Tesla MRI is 14.75% (95% CI 8.25-24.97).

Discussion: Overall, the global prevalence of radiologically identifiable MCDs is substantial and nearly doubles in people with DRE. Poor maternal health, exposure to toxins, pollution, radiation and immune activation are risk factors for MCDs. The lower prevalence in LMICs may be related to differences in social determinants of health, including inequities in access to health care and the lack of sophisticated diagnostic facilities. Premature mortality in those who have MCDs cannot be excluded. The lower prevalence observed in LMICs likely reflects disparities in healthcare access and higher premature mortality rates rather than lower incidence.

76

Thalamic Volumetric Analysis in ESES using THOMAS: An MRI Case-Control Study

Dion Tahiri^1,2, Suresh Pujar^2,3, Aswin Chari^1,4

¹University College London , ²Developmental Neurosciences, Institute of Child Health, Great Ormond Street, ³Department of Neurology, Great Ormond Street Hospital, ⁴Department of Neurosurgery, Great Ormond Street Hospital

Introduction: Electrical status sleep epilepticus (ESES) is the terminology used to describe an epileptic pattern observed during NREM sleep of paediatric patients who present with developmental and behavioural deficits. These deficits are induced by a chronic form of NREM sleep epilepsy that can be described under an umbrella of different terms such as DEE-SWAS or CWSW. Previous studies have suggested that early lesions or developmental abnormalities of the thalami on either hemisphere may lead to the onset of ESES pathology, while others simply outline the correlation between a seemingly reduced thalamic volume and occurrence of this wave pattern; results that have been inconsistent across the literature.

Methods: We conducted a retrospective case control study of 29 ESES patients who were compared to 65 healthy controls, using post-diagnostic MRI files obtained between 2013-2022. A converted form of these files was then run through THOMAS, a docker-based container that automatically segments, labels, and measures the volumes of the thalamus and its constituents in both the left and right hemispheres. We had run various age and gender-adjusted statistical tests, including total volume comparison, individual abnormality testing using Z scores, lateralised atrophy index testing, using MATLAB.

Results: Our age and gender adjusted ANCOVA test had found that the ESES cohort had a reduced thalamic volume (-1805mm3, Cohen’s D = 1.67, p < 0.001) at a group level. Individually, 59% of ESES patients fell outside of the 95th percentile of normal volume (Z < -1.96) 8 of 11 nuclei showed significant volume decrease after Bonferroni correction, the biggest decreases being the mediodorsal-parafasicular (d=1.56), ventral lateral posterior (d= 1.56), and ventral anterior nuclei (1.52). Our lateralisation testing had showed significantly increased thalamic asymmetry in the ESES cohort (p=0.034), not be significantly linked to a specific hemisphere (sign test p = 0.15).

Discussion: Our data supports the idea seen across the literature that a reduction in thalamic volume occurs within patients with ESES, yet whether this is the cause of the disorder or an effect caused by some other factor remains unknown. Our results may lead to future thalamic studies that highlight the pathology of this condition, the way both the reduced overall thalamic volume, and the hemispheric thalamus imbalance correlate with disease severity, seizure onset, and the extent of the ESES waveform during NREM sleep.

77

New-Onset Refractory Status Epilepticus (NORSE) versus Refractory Status Epilepticus not meeting NORSE criteria: A Comparative Clinical and EEG-Based Study

Seren Hawksworth¹, Nina Moutonnet^2,3, Gregory Scott², Sanjeev Rajakulendran¹, Mahinda Yogarajah¹, Matthew Walker¹, Umesh Vivekananda¹

¹UCL Queen Square Institute of Neurology, ²Imperial College London, UK Dementia Research Institute, ³Imperial College London, Department of Computing

Background: New-onset refractory status epilepticus (NORSE) is a rare, severe presentation of refractory status epilepticus (RSE), with approximately half of cases cryptogenic (c-NORSE). We compared electroencephalography (EEG) findings alongside clinical features between NORSE and RSE not meeting NORSE criteria to better understand outcomes and prognostication.

Methods: This retrospective cohort study analysed 81 RSE patients admitted to the Intensive Therapy Unit (ITU) at The National Hospital for Neurology and Neurosurgery and University College London Hospital (2007-2023). Patients were categorised as RSE not meeting NORSE criteria (n=52) or NORSE (n=29). EEG features, treatment, and outcomes (ITU stay, mortality, cognitive outcome, and anti-seizure medication (ASMs) use), were evaluated.

Results: Of 29 NORSE patients (15 c-NORSE, 51.7%; 14 aetiology-identified NORSE, 48.3%), autoimmune (20.7%) and infectious (17.2%) aetiologies were most prevalent. Six-month mortality was higher in c-NORSE than aetiology-identified NORSE (20.0% vs. 7.1%), but similar between NORSE and RSE not meeting NORSE criteria (13.8% vs. 21.1%). EEG analysis revealed longer seizure duration (mean 243.38 vs. 51.67 seconds; p<0.001), a tendency to higher seizure frequency (0.013 vs. 0.009 seizures/minute; p=0.076), and increased presence of generalised periodic epileptiform discharges and stimulus-induced rhythmic, periodic, or ictal discharges in NORSE than RSE not meeting NORSE criteria. Across all patients, higher initial seizure burden correlated with longer ITU admissions and worse outcomes. Compared with RSE not meeting NORSE criteria, NORSE patients had longer ITU stays (median 51 vs. 11 days; p<0.001), more ASMs (mean 3.91 vs. 2.00; p=0.031), greater sedative use (mean 2.17 vs. 1.00; p<0.001), and more frequent cognitive impairment at discharge (88.9% vs. 62.2%; p=0.037).

Conclusions: NORSE, particularly c-NORSE, tends to follow a more severe course than RSE not meeting NORSE criteria, with more severe background electroencephalographic features including the initial EEG, greater seizure burden, longer ITU stays, and increased incidence of cognitive impairment at discharge. Six-month mortality was similar, implying additional contributing factors to mortality.

78

Validating SEEG Ictal and Interictal Pattern Classification

Li-Zhang Tan¹, Khalid Hamandi², Glen Brimble²

¹Cardiff University, ²Cardiff and Vale University Hospital of Wales

Aim: To identify seizure onset patterns (SOP) in patients diagnosed with hippocampal sclerosis (HS) and focal cortical dysplasia (FCD) using stereo-electroencephalography (SEEG), and to compare findings with those reported by Di Giacomo et al.

Methods: Data from 13 patients with drug-resistant focal epilepsy were retrospectively collected from the University Hospital of Wales. Diagnosis of HS and FCD were established through clinical presentation, neuroimaging, and later concluded with SEEG findings. Ictal and interictal SEEG recordings were reviewed, and SOPs were classified according to Di Giacomo’s categories: low voltage fast activity (LVFA), fast activity (FA), rhythmic sharp activity (RSA), repetitive fast spike burst (RFSB), and slow burst (SB), based on the earliest identifiable ictal changes.

Results: In total, 153 clinical seizures (99 HS, 54 FCD) and 29 electrographic seizures (22 HS, 7 FCD) were reviewed across 13 patients (9 HS, 4 FCD). In HS, LVFA was the predominant SOP (68.7%), followed by RSA (30.3%) and RFSB (1%). FCD seizures showed a more heterogeneous distribution, with RSA and RFSB each accounting for 29.6%, followed by LVFA (20.4%) and SB (20.4%). At the patient level, LVFA was present in all HS patients (100%) and 75% of FCD patients. LVFA demonstrated high sensitivity (100%) but low specificity (25%) for HS, while RFSB showed low sensitivity (25%) but higher specificity (88.9%). Fischer’s exact testing revealed no statistically significant association between any individual SOP and underlying pathology. Similar patterns were observed in electrographic seizures, with no single onset pattern reliably distinguishing HS from FCD.

Conclusion: Seizure onset patterns demonstrated substantial overlap between HS and FCD in this cohort. Although LVFA was universally observed in HS and certain patterns appeared more frequently in FCD, no single onset morphology achieved sufficient sensitivity or specificity to reliably distinguish the two pathologies. These findings contrast with those of Di Giacomo et al., suggesting that seizure onset morphology alone has limited diagnostic utility. Given that the neurophysiological mechanisms underlying this discharge patterns remain incompletely understood, establishing a direct correlation between ictal onset pattern and histopathology is inherently challenging. Electrode locations also influence recorded SEEG patterns. Therefore, the reproducibility and generalisability of previously described SEEG patterns across different centres and patient population remain important considerations. These findings reinforce the need for comprehensive multimodal evaluation in presurgical assessment rather than reliance on seizure onset morphology alone.

79

Clinical decision support after inconclusive EEG: evaluating the clinical utility and health economic potential of BioEP

Phil Tittensor^1,4, Anke Verhaege², Jeremy Andrews², Wessel Woldman², John Terry^2,3, Daniel Russell¹, Elizabeth Schnabel¹, Conor Smyth¹, Jacqui Rowe¹, Toni Ball¹, Hannah Johnson¹, Louise Coope, Francesco Manfredonia¹

¹Royal Wolverhampton NHS Trust, ²Neuronostics, ³University of Plymouth, ⁴University of Wolverhampton

Background: BioEP is a clinical decision-support tool designed to aid differentiation between epilepsy and non-epileptic conditions following an inconclusive electroencephalogram (EEG). The tool utilises computational biomarkers derived from routine EEG background activity, which are fed into a statistical model to output one of five ordinal classes: very unsupportive, unsupportive, neutral, supportive and very supportive of epilepsy. Inconclusive EEG results are common and can contribute to diagnostic uncertainty, the need for repeat investigations and delays in diagnosis and treatment, increasing healthcare resource use and costs. This first real-world evaluation at The Royal Wolverhampton NHS Trust assesses the potential health economic value of BioEP as well as the adoption and clinical utility of the tool in routine practice within a nurse-led seizure clinic.

Methods: Interim data were reviewed for all EEGs uploaded to BioEP up to 24 June 2026, 6 months after implementation. Four senior Epilepsy Specialist Nurses (ESNs) requested and reviewed BioEP assessments. Clinical utility and potential health economic impact were assessed using clinician-reported data collected via a structured questionnaire, including effects on diagnostic confidence, downstream clinical activity, and perceived value during patient discussions.

Results: At the interim evaluation, 96 BioEP assessments had been requested, and clinician-reported evaluation data were available for 66 cases. Of the 96 assessments, 8 (8.3%) were classified as very supportive, 12 (12.5%) supportive, 40 (41.7%) neutral, 20 (20.8%) unsupportive, and 16 (16.7%) very unsupportive of epilepsy. Use of BioEP increased during the implementation period and then remained consistent from month 3 onward. BioEP improved clinician confidence in diagnosing epilepsy in 39 of 66 evaluated cases (59%). No additional EEG investigations were requested in 37 of these 39 cases (94.8%). In cases where clinician confidence was not improved by BioEP; no further testing was requested in 23 of 27 cases (85,1%). BioEP findings were discussed with patients in 28 of 66 cases (42.4%) and were considered by clinicians to add value in 19 of those 28 discussions (67.8%).

Conclusion: BioEP was associated with improved clinician confidence, which may lead to more efficient use of clinical resources. Furthermore, the interim data confirm that BioEP can be integrated into routine clinical practice after an initial learning period, as its use increased over the first few months before stabilising. The strength of the current results is limited by the limited sample size. Ongoing data collection aims to provide further evidence regarding the impact and scalability of BioEP in routine practice.

80

X-SEPT: individualised, EEG based overnight music playlists for people with drug-resistant seizures – a pilot study

Phil Tittensor^1,3, Nigel Osbourne², Kit Barnes³, Rahnuma Feist-Hassan³, James Hanlon³, John Turner³, Jonathan Walton³

¹Royal Wolverhampton NHS Trust, ²University of Edinburgh, ³University of Wolverhampton, 4X System

Purpose: Up to 70% of people with epilepsy (PWE) can become seizure free using anti-seizure medication (ASM). The remaining 30% require alternative management strategies. Listening to Mozart’s K448 has been reported to reduce seizure frequency for some PWE. X-SEPT produces playlists created through a personalised process, selecting music tracks based primarily on musical feature similarity to a ‘healthy’ period of the patient’s EEG. This has been audified with the aim of encouraging healthy brain activity during sleep through neural synchronisation/entrainment. The tracks are edited to reduce sudden dynamic shifts or uncomfortable pitches. A previous exploratory study reported a mean 28.42% reduction in epileptiform spikes.

Objective: We will evaluate the efficacy, safety, and cost-effectiveness of using X-SEPT music playlists nightly by people with drug resistant epilepsy, through assessing seizure frequency, seizure severity, seizure safety, quality of sleep, quality of life and cost data, compared to listening to a nightly playlist of Mozart’s K448 or to receiving care as usual in a three arm parallel group randomised controlled trial with follow-up phase; an internal pilot study.

The secondary objective is to assess the study design in terms of participant recruitment rates, retention rates, protocol adherence, and data completion, so that any design corrections can be made (if necessary) before a larger RCT is carried out.

Method: Forty-eight PWE receiving care at The Royal Wolverhampton NHS Trust will be recruited. The study will have two phases; confirmatory phase consisting of baseline (1 month) and intervention (3 months) periods, followed by the surveillance phase that will consist of an optional follow-up period (12 months).

Participants will be randomised into an intervention group, where they will receive the X-SEPT playlist, an alternate group where they will receive Mozart K448 on loop, or a control group where they will receive care as usual.

Data will be collected through forms, seizure diaries, sleep questionnaires, saliva samples, SUDEP checklists, cost questionnaires, and quality-of-life questionnaires.

Results: Preliminary results from the first phase will be presented at conference.

Conclusion: Completion of the pilot study should support progression from an exploratory study to a full randomised controlled trial of X-SEPT in drug-resistant epilepsy. If shown to be effective, X-SEPT may offer a low-risk, non-invasive adjunctive approach for people with drug-resistant seizures.

81

From algorithm to patient: progress from the ATMOSPHERE (Artificial intelligence To Optimise Seizure Prediction to Empower people with Epilepsy) project

Phil Tittensor^1,2, Amberly Brigden³, Leandro Junges⁴, Peter Kissack⁴, Yasser Qureshi⁶, Emily Quilter³, Samuel Downes³, Liz Stuart⁷, Rosie Charles⁵, Amelia Slay³, Jerussa Vasikaran³, Lauren Thompson³, Emily Nielsen³, Matthew Wragg³

¹Royal Wolverhampton NHS Trust, ²University of Wolverhampton, ³University of Bristol, ⁴University of Birmingham, ⁵Neuronostics, ⁶University of Warwick, ⁷Ulster University

Purpose: Seizure unpredictability is a research priority for people living with epilepsy (PWE). ATMOSPHERE is an interdisciplinary project to design technology that alerts individuals if seizures are likely in the near future; analogous to checking a weather forecast for the need of an umbrella.

Method: Synthetic seizure diaries and publicly available datasets were used to train an AI algorithm to forecast seizures. 22 stakeholders (clinicians, individuals with lived experience) co-designed the smartwatch and smartphone device, designed to capture the real patient data needed to optimise the AI algorithm. Stakeholders advised about the optimal design of presenting these forecasts, including an artist’s interpretation using video and pictures to explain the project. Subsequently, 14 participants used the device for up to 3 months in their lived context, providing longitudinal data for the seizureforecasting algorithm. A small subset (n=4) are using long-term implantable EEG (UNEEG Episight), with the aim of enabling accurate mapping of the forecast to seizures. The project now has ethical approval to begin a feasibility study of the device within an NHS clinic.

Results: We will present:

The algorithm pipeline based on synthetic dataset and refined in real patient data

Usability of the smartwatch and smartphone device

The optimal design of how to present forecasts, including artist designs for public engagement forums

UNEEG analysis

Usability withing NHS feasibility study

Conclusion: ATMOSPHERE is a novel concept, utilising readily available technology to forecast seizures. This has the potential to minimise risk, up to and including SUDEP, and maximise opportunities for people with epilepsy. Our results so far demonstrate high user acceptability and an ability to record the necessary data to produce a forecast. We will begin a trial to produce forecasts for people with epilepsy in 2026 and hope to be able to present preliminary results at the ILAE Scientific Meeting.

82

Preoperative Quantitative EEG Predicts Long-Term Tau Accumulation and Cognitive Decline Following Temporal Lobe Epilepsy Surgery

Adolfo Mazzeo^1,2, Thaera Arafat¹, Luisa Delazer¹, Isha Puntambekar¹, Kjell Erlandsson⁴, Beate Diehl^1,3, Dr Fahmida Chowdhury^1,3, Alicija Mrzyglod¹, Maria Thom¹, Matthias Koepp^1,3

¹UCL Queen Square Institute of Neurology, ²Department of Human Neurosciences, Sapienza University of Rome, ³National Hospital of Neurology and Neurosurgery, ⁴Institute of Nuclear Medicine, University College London Hospital

Background: Chronic network hyperexcitability has emerged as a potential driver of tau accumulation and accelerated brain ageing in temporal lobe epilepsy (TLE), yet electrophysiological markers identifying networks susceptible to activity-dependent tau propagation remain unknown. We investigated whether preoperative quantitative EEG (qEEG) measures are associated with postoperative tau burden measured approximately one decade later and with long-term cognitive outcome.

Methods: Patients with drug-resistant TLE underwent tau PET imaging with [¹⁸F]MK-6240 following epilepsy surgery. After preprocessing, routine preoperative EEG recordings were analysed with exact low-resolution electromagnetic tomography (eLORETA) source localisation. Regional source activity was quantified across conventional frequency bands and correlated with global and regional tau PET uptake after adjustment for age. Associations with phosphorylated tau (AT8) in surgical tissue, long-term cognitive decline and postoperative seizure outcome were also explored.

Results: Preoperative qEEG demonstrated significant associations with subsequent tau burden. Lower frontal (ρ = −0.544, p = 0.029) and central delta activity (ρ = −0.552, p = 0.027) was associated with higher global tau PET uptake, together with higher temporal alpha2 (ρ = 0.542, p = 0.030) and alpha3 activity (ρ = 0.582, p = 0.018). Similar relationships were observed for temporal neocortical tau uptake. The same qEEG pattern predicted long-term memory decline, with patients exhibiting post-operative cognitive deterioration showing lower frontal and central delta activity and higher temporal alpha2 activity. In contrast, qEEG measures were unrelated to phosphorylated tau burden within the resected tissue or postoperative seizure outcome.

Conclusions: Preoperative qEEG source activity identifies network-level physiological states associated with subsequent distributed tau accumulation and long-term cognitive decline. The dissociation between qEEG and pathological tau within the surgical specimen suggests that electrophysiological network function reflects vulnerability to future activity-dependent tau propagation rather than existing focal tau pathology. Together with recent evidence linking impaired glutamate homeostasis to chronic network hyperexcitability, these findings support a model in which preserved but chronically hyperactive neural networks promote progressive tau-associated brain ageing before overt network failure becomes clinically apparent.

83

Long-term EEG - Rapid Treatment Changes (LEG-RTC): a novel use of implantable EEG in a non-tertiary clinical setting

Phil Tittensor^1,2, James Barraclough¹, Neil McNiven¹, Francesco Manfredonia¹, Daniel Konn¹, Kim Byrne³, Pia Kjaer Gauger³, Chris Young³

¹Royal Wolverhampton NHS Trust, ²University of Wolverhampton, ³UNEEG Medical

Purpose: To assess whether rapid changes of treatment could be undertaken safely with continuous EEG monitoring for people with difficult to control seizures.

Method: Five patients were selected from the caseload of the epilepsy nursing service at the Royal Wolverhampton NHS Trust (RWT). All patients reported frequent seizures, at least monthly, and had tried at least four different antiseizure medications (ASM). A long-term, two channel EEG (UNEEG) was implanted in each patient by an ENT surgeon between October 2025 and January 2026. Patients retained the device for six months from implantation. Patients had EEG reports prepared by UNEEG every two weeks, ratified by consultant neuro physiologists from RWT. Patients were remotely reviewed by a consultant nurse for the epilepsies every two weeks, with treatment changes made based upon EEG recorded events.

Results: Five patients were successfully implanted. One patient developed an infection around the implantation site and had to be explanted. The four remaining patients wore the device for more than 90% of the time. One patient was aware of every seizure. Two patients reported events they considered to be seizures without EEG correlation; both had other seizures recorded that they were unaware of. One patient recorded no seizures for five months, after which there were two clusters of seizures. All patients undertook rapid changes to antiseizure medication, reacting to recorded seizures. One patient showed significant improvements in seizure control two reported improved understanding of their seizures. Two patients UNEEG recordings showed excessive sleep, which has led to discussions about medication changes not purely based on seizure control.

Conclusion: We present the first experience of UNEEG implanted by non-neurosurgeons outside a tertiary epilepsy centre, with clinical consultations undertaken by a consultant nurse. Our experience demonstrates the potential benefit of this technology for patients who are not managed in tertiary settings.

84

Current availability and demand for Point-of Care EEG in the across Critical Care in the United Kingdom: A National Survey

Masumi Tanaka¹, Dr Fahmida Chowdhury², Helen Ford¹, Hannah Stapley³, Juliet Solomon³, Khalid Hamandi⁴, Tejal Mitchell⁵, Tim Fudge⁶, Rajiv Mohanraj⁶

¹Kings College Hospital, ²National Hospital for Neurology and Neurosurgery, UCLH, ³ILAE British chapter, ⁴Cardiff and Vale University Health Board, Cardiff University, ⁵Cambridge University Hospitals NHS Foundation Trust, ⁶Manchester Centre for Clinical Neurosciences, Salford Royal

There is growing evidence that point-of-care EEG (POC-EEG) may be useful or associated with improved outcomes in critical care patients in status epilepticus, subarachnoid haemorrhage and reduced consciousness of unknown aetiology. In the UK, the use and demand of POCEEG remains unknown. We conducted a national survey to determine the current use and demand of POC=EEG.

Method: A 17-question survey, subcategorised into demographics of the responder and their hospital, current provision of departmental EEG and POCEEG, and the respondents’ views on POCEEG usage in critical care, including a free comments question, was distributed electronically to members of the Neuroanaesthesia and Critical Care Society (NACCS), Faculty of Intensive Care Medicine (FICM), Advanced Critical Care Practitioner (ACCPs) and the British Association of Critical Care Nursing (BACCN) network between December 2025 and April 2026, reaching 250 intensive care units across the UK. Respondents included senior nurses, advanced critical care practitioners (ACCPs) and consultants. Where multiple responses were received from the same hospital regarding EEG provision, the most senior respondent was used.

Results: Respondents: A total of 115 responses were received (30 ACCPs, 79 consultants, and 6 senior nurses). Three types of ICUs were represented - 47 non-neuroscience ICUs (38 of which were general ICUs), 12 general ICUs managing neurocritical care patients, and 7 dedicated neuro ICUs). 10% (7/66) of hospitals have either reduced or full montage on-demand EEG, most of which are centres with neurocritical care patients (6/7). Access to departmental EEG ranged from none to 7 days, with 68% of hospitals having 5 day access. Reporting: EEGs were mostly reported by neurophysiologists (79%) and neurologists (15%) of cases. Demand: When asked if respondents would like to have better access to continuous bedside EEG (N=115), 59% said yes, 16% said maybe and 12% said no. 38% believed that better access to EEG could shorten the duration of mechanical ventilation for some patients, 33% thought that it may do, 14% were not sure whereas 3% did not believe so. Free text responses consistently highlighted the need for more efficient EEG access, availability and reporting to support decision making at the bedside.

Conclusions: A significant proportion of intensivists perceived POC-EEG to be of benefit. Our survey highlights the low availability of POC-EEG in ICUs in the UK, and the need to develop infrastructure for POC-EEG guided ICU therapy at the bedside.

85

Prioritising follow-up for people with suspected epilepsy using a digital EEG biomarker

Wessel Woldman, Rosie Charles, Emanuela De Falco, Elizabeth Galizia, David Martin-Lopez, Kay Meiklejohn, David Allen, Rohit Shankar, John Terry

Neuronostics

Introduction: Diagnostic delays for suspected epilepsy remain a critical challenge. With initial routine EEGs proving non-contributory in over 70% of cases, people with suspected epilepsy can wait several months for follow-up testing. This creates substantial risks; for example, 35% experience further seizures within twelve months whilst awaiting diagnosis. We investigated whether a digital EEG biomarker could help prioritise those most likely to have epilepsy for expedited testing.

Methods: We analysed 196 non-contributory routine EEGs from six NHS sites (N=66 ultimately diagnosed with epilepsy; N=130 with alternative conditions). BioEP, a digital biomarker consisting of eight computational features (including spectral properties, network connectivity, and model-based measures) that gives a measure of the level of support for epilepsy, was calculated for each EEG. We simulated clinical prioritisation by comparing biomarker-ordered follow-up lists against standard scheduling (i.e. ordered by time of referral) across random patient subsets (N=50; stratified random sampling).

Results: Biomarker-based prioritisation consistently outperformed standard scheduling. To see 50% of epilepsy patients required significantly fewer follow-up EEGs (median 33% vs 49% of list; p<0.001). The diagnostic yield for epilepsy at the 50% follow-up mark increased from 27% with random scheduling to 33% with biomarker ordering (p<0.001). The proportion of epilepsy patients seen after 50% follow-up visits increased by 10% (median value, 95% CI 7-15%, effect size 0.76). Benefits were most pronounced in typical clinic scenarios where epilepsy prevalence was 20-50% of the total list.

Discussion: This multicentre study demonstrates that computational analysis of non-contributory EEGs can identify patients at higher risk of epilepsy, enabling more efficient resource allocation. Prioritising those most likely to benefit from second-line investigations can potentially reduce diagnostic delays and seizure burden whilst maintaining equitable access, as no one is excluded from follow-up. A prospective two-arm study (‘PRIORITISE’) is currently underway to prospectively assess the potential of this methodology.

The author-list was condensed: full author-list at: DOI: 10.1016/j.yebeh.2026.110925

86

A novel computational EEG biomarker to support diagnosis in paediatric epilepsy: results from the OASIS study

Wessel Woldman, Emanuela De Falco, Verity Fuller, Kay Meiklejohn, Kelly St. Pier, John Terry

Neuronostics

Background: Epilepsy is the most common long-term neurological disorder in childhood, yet diagnosis remains challenging because routine EEG has limited sensitivity, particularly when epileptiform abnormalities are absent. Recently, a computational EEG biomarker of epilepsy based on network science has demonstrated robust diagnostic performance in adults using non-contributory EEG recordings (Tait et al., 2024) and was translated into a clinical decision-support tool (BioEP). However, because this biomarker was developed in adult populations, it cannot be directly applied to paediatric EEG without accounting for fundamental differences in brain maturation, developmental network organisation, and paediatric comorbidities.

Methods: We conducted a retrospective multicentre study (OASIS) of routine non-contributory EEG recordings from children investigated for suspected epilepsy across seven NHS sites. The final cohort comprised 530 patients (ages: 2-17 years, N=261 epilepsy, N=269 differential). Biomarker development was performed on a stratified training subset (n=424), with an established set of computational EEG features extracted from each recording.

We first assessed the influence of potential confounders, including age and comorbidities. Age-subgroup analyses were performed across three clinically pre-defined age groups (2-4, 5-9, and 10-17, Meiklejohn et al. 2025). We then developed a model framework using nested cross-validation for model selection and hyperparameter optimisation to identify the best-performing model.

Results: We found that age had a substantial effect on several computational EEG features. Comparison with an adult cohort showed that these age-related trends extended into adulthood. Notably, age influenced not only the feature values but also their variability, which increased with age, highlighting the dynamic nature of EEG network organisation across development and reinforcing the need for age-aware modelling.

Despite this developmental variability, the EEG features retained diagnostic value. A tree-based classifier incorporating EEG features and demographic confounders (age, sex, comorbidity, and ASM status) achieved performance significantly above chance (AUC: 73.3%, balanced accuracy: 70%, specificity: 75%, sensitivity: 65%) and consistent across the age groups, demonstrating discrimination between epilepsy and non-epilepsy beyond what achievable with conventional EEG alone.

Conclusions: Developmental changes substantially influence computational EEG features in children and should be considered when developing diagnostic tools for paediatric epilepsy. Age-adjusted computational models can identify epilepsy with clinically meaningful accuracy even from clinically inconclusive EEGs. These findings support computational EEG biomarkers as an additional source of diagnostic evidence alongside routine clinical EEG interpretation that may contribute to earlier and more accurate diagnosis in children with suspected epilepsy.