Affordable EEG, Actionable Insights: An Open Dataset and Evaluation Framework for Epilepsy Patient Stratification

HM Shadman Tabib, Md. Hasnaen Adil, Ayesha Rahman, Ahmmad Nur Swapnil, Maoyejatun Hasana, Ahmed Hossain Chowdhury, A. B. M. Alim Al Islam · 2025

Read Affordable EEG, Actionable Insights: An Open Dataset and Evaluation Framework for Epilepsy Patient Stratification by HM Shadman Tabib, Md. Hasnaen Adil, Ayesha Rahman, Ahmmad Nur Swapnil, Maoyejatun Hasana, Ahmed Hossain Chowdhury, A. B. M. Alim Al Islam (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-01879.txt.

Introduction

HM Shadman Tabib and six colleagues take epilepsy care where clinical brain-wave gear cannot go. Multi-channel EEG is scarce in many regions. So the team collected NEUROSKY-EPI. It is the first open dataset of single-channel, consumer-grade EEG for epilepsy. The recordings come from a South Asian clinic, with rich notes on each patient.

To show the data's worth, they built EmbedCluster. The pipeline borrows features from EEGNet models trained on clinical data. It adds features from a contextual autoencoder. Then it groups patients by their EEG patterns, with no labels. Even cheap one-channel signals sorted patients in meaningful ways. The authors stress human concerns too: use in poor settings, clarity for non-specialists, and guards for privacy and bias. The dataset and code are open.

Catalog record

CREATORHM Shadman Tabib, Md. Hasnaen Adil, Ayesha Rahman, Ahmmad Nur Swapnil, Maoyejatun Hasana, Ahmed Hossain Chowdhury, A. B. M. Alim Al Islam
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE50 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-01879

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Affordable EEG, Actionable Insights: An Open Dataset and Evaluation Framework for Epilepsy Patient Stratification HM Shadman Tabib, Md. Hasnaen Adil, Ayesha Rahman, Ahmmad Nur Swapnil, Maoyejatun Hasana, Ahmed Hossain Chowdhury, A. B. M. Alim Al Islam Abstract . Access to clinical multi-channel EEG remains scarce in many regions worldwide. We present NEUROSKY–EPI , the first open dataset of single-channel, consumer-grade EEG for epilepsy, collected in a South Asian clinic together with rich contextual metadata. To probe its utility, we introduce EmbedCluster , a patient-stratification pipeline that transfers representations from EEGNet trained on clinical recordings and augments them with contextual autoencoder embeddings, before applying unsupervised clustering to group patients by EEG patterns. Results indicate that low-cost, single-channel data can indeed support meaningful stratification. Beyond algorithmic performance, we foreground CHI concerns: deployability in resource-constrained settings, integration into everyday workflows, interpretability for non-specialists, and safeguards for privacy, inclusivity, and bias. By releasing the dataset and code, we aim to catalyze interdisciplinary work across HCI, health technology, and machine learning—charting a path toward affordable, actionable EEG that supports equitable epilepsy care 1. Introduction Building on this vision, we now situate our work in the broader context of diagnosis and monitoring of epilepsy. Epilepsy is a widespread neurological disorder that affects an estimated 50 million people worldwide ( WHO2019Epilepsy ) . Continuous electroencephalography (EEG) remains the gold standard in clinical diagnosis and surveillance of seizures, but access to multichannel EEG systems is severely limited in many low and middle income regions due to high equipment costs, infrastructure demands, and the need for trained specialists ( Tatum2018EEGUtility ) . These barriers often result in delayed diagnosis, irregular follow-up, and reduced access to effective treatment. Addressing this gap requires new approaches that balance clinical utility with accessibility. We therefore ask: What would epilepsy care look like if every rural clinic had a $100 EEG stratifier? Recent work in human-computer interaction (HCI) has highlighted the growing potential of consumer-grade wearable EEG technologies to bring brain sensing to everyday contexts ( Casson2010WearableEEG ; Patel2012WearableEEG ) . Dry-electrode headsets integrated into familiar form factors, such as headphones, caps, or earbuds, reduce setup time, improve comfort, and enable data collection outside of traditional clinical environments. This shift opens opportunities for biosensing applications beyond the lab, including neurofeedback, interaction design, and health monitoring ( Zander2010TowardsPassive ; Kosmyna2019AttentionAugmentation ) . In particular, the integration of EEG into consumer devices supports human-centered goals of accessibility, usability, and inclusivity. For epilepsy specifically, early feasibility studies suggest that low-cost, single-channel EEG headsets can capture physiologically meaningful markers of abnormal brain activity ( Casson2017SingleChannelEEG ; NeuroSkyApplications ) . Portable systems like the NeuroSky MindWave have been validated against clinical-grade EEG for tasks such as attention and meditation monitoring, demonstrating that consumer hardware can approximate traditional measurements ( NeuroSkyApplications ) . More recent systematic reviews highlight the growing clinical interest in mobile EEG for seizure monitoring and management, suggesting that such approaches could complement or extend conventional hospital-based EEG systems ( Biondi2022MobileEEGReview ) . Despite these encouraging developments, significant gaps remain. First, most existing epilepsy datasets rely on multi-channel, hospital-grade EEG collected in controlled laboratory or clinical environments ( Goldberger2000PhysioNet ) . Open datasets using consumer-grade, portable devices are nearly absent, creating barriers for researchers and designers interested in studying real-world, low-cost seizure monitoring. Second, many current machine learning approaches are data-intensive and assume large volumes of clean, multi-channel input. These assumptions limit generalizability and neglect the contextual factors—such as whether a patient is resting or alert—that can strongly influence EEG signals ( Craik2019DeepEEGReview ; Roy2019DeepEEGSurvey ; Lawhern2018EEGNet ) . To address these challenges, we introduce NEUROSKY–EPI , to our knowledge the first open-access epilepsy EEG dataset collected using a single-channel consumer-grade headset in a South Asian low-resource context. We further propose EmbedCluster , an unsupervised pipeline that combines EEGNet embeddings trained on clinical data with contextual autoencoder embeddings, then applies clustering to stratify patients by EEG patterns. Our contributions are as follows: C.1 Open Dataset: NEUROSKY–EPI, comprising 25 patients and over 2,000 labeled windows, with rich metadata including seizure types, medications, and demographics. C.2 Method: An unsupervised clustering framework that extracts deep neural embeddings from EEGNet and complements them with contextual autoencoder embeddings. C.3 Evaluation: Empirical demonstration of 62.50% clustering accuracy with EEGNet embeddings and comparable performance (58.33–62.50%) with contextual autoencoders, both exceeding chance (50%). C.4 HCI Implications: Reflection on usability, deployment, and ethics, highlighting how affordable EEG-based clustering could enable more inclusive epilepsy risk assessment in underserved communities. Together, these contributions bridge machine learning rigor with HCI design concerns, laying the groundwork for future interdisciplinary work at the intersection of ubiquitous computing, health technology, and equitable care through intelligent patient stratification. To position these contributions within existing scholarship, we next review prior work on clinical EEG, consumer-grade alternatives, machine learning approaches, and open datasets. 2. Related Work Clinical EEG and Epilepsy Monitoring. Electroencephalography remains the definitive diagnostic tool for epilepsy, with decades of research establishing its clinical utility ( Tatum2018EEGUtility ; WHO2019Epilepsy ) . Traditional multi-channel EEG systems, following the international 10–20 electrode placement system, capture complex spatiotemporal seizure patterns ( Goldberger2000PhysioNet ) . However, these systems demand specialized training, controlled environments, and significant infrastructure ( Biondi2022MobileEEGReview ) . While continuous EEG monitoring can improve detection rates by 50–80% compared to routine short-term recordings ( Tatum2018EEGUtility ) , access remains limited in low-resource settings. WHO estimates that up to 75% of people with epilepsy in developing countries do not receive appropriate treatment, partly due to diagnostic limitations ( WHO2019Epilepsy ) . This gap has motivated exploration of alternative monitoring approaches that balance diagnostic capability with accessibility and cost-effectiveness ( Casson2010WearableEEG ; Patel2012WearableEEG ) . Wearable and Consumer EEG Technologies. The emergence of dry-electrode EEG has enabled portable brain monitoring ( Casson2017SingleChannelEEG ; Casson2010WearableEEG ; Chi2010DryContact ) . Unlike wet electrodes, dry sensors can be embedded into everyday objects (headphones, caps, earbuds), reducing setup time and improving comfort ( Fiedler2015NovelEEG ; Wang2016DryElectrode ) . Ambulatory systems such as smartphone-compatible ear electrodes ( Bleichner2016ConcealedEEG ) achieve comparable signal quality to traditional scalp recordings. Consumer devices like the NeuroSky MindWave, Emotiv EPOC, and Muse headband have further broadened access ( Hairston2014UsabilityWearable ) . Advanced systems now include wireless streaming ( Lin2010WirelessEEG ) , real-time processing ( Mullen2015RealTimeEEG ) , and flexible electronics ( Rogers2010FlexibleWireless ; Norton2015SoftWireless ) . HCI researchers have investigated EEG in ubiquitous computing environments ( Zander2010TowardsPassive ; Kosmyna2019AttentionAugmentation ) . Studies suggest that consumer EEG devices can often be operated with minimal training, enabling self-administered health monitoring ( Patel2012WearableEEG ; Xu2017ReviewPortable ) . Yet limitations remain: higher noise, lower spatial resolution, and variable reliability compared to clinical systems ( Raduntz2017EEGArtifact ) . Still, validation work shows these devices can capture meaningful physiological markers (attention, meditation, basic neurological assessment) ( NeuroSkyApplications ; Aspinall2006PortableEEG ) . In epilepsy, mobile EEG offers shorter setup times, more naturalistic data collection, and earlier seizure detection ( Biondi2022MobileEEGReview ; Ramgopal2014SeizureDetection ) . These advances highlight feasibility but underscore the need for datasets beyond controlled clinical environments. Machine Learning for Seizure Detection. Automated seizure detection has progressed from hand-crafted features (spectral power ( Liang2010CombinationFeature ) , entropy ( Kannathal2005EntropiesEEG ) , wavelets ( Adeli2003WaveletEEG ; Faust2015WaveletBased ) , statistical moments ( Subasi2005ClassificationEMG )

Read the complete text