Epilepsy affects millions of people worldwide, and one of its most debilitating aspects is the unpredictability of seizures. Advances in machine learning and neurotechnology offer new opportunities to tackle this challenge. This competition invites researchers, data scientists, and clinicians to develop robust algorithms capable of predicting seizures using ultra long-term (months) EEG recordings with a subcutaneous EEG device.
 
Objective
 
Participants will create algorithms to predict seizure occurrences based on extended EEG covering several months of almost continuous subcutaneous EEG recordings in a real life setting. The focus is on detecting pre-ictal states—specific patterns or biomarkers indicating an impending seizure—while minimizing false positives to ensure practical utility.
 
Dataset
 
The competition dataset comprises:
●    A dataset containing ultra long-term EEG recordings from three patients using a 24/7 EEG SubQ device (UNEEG medical A/S, Lynge, Denmark) is provided for the challenge. At least 25 lead seizures per patient are included in the training data.
●    Annotations marking seizures. 
●    An independent test dataset is used to evaluate submitted algorithms
 
Challenge Goals

●    Sensitivity: High detection rate of true pre-ictal periods.
●    Specificity: Minimizing false alarms to avoid undue burden on patients.
●    Scalability: Algorithms must handle ultra long-term recordings efficiently.
●    Generalizability: Solutions should perform well across diverse patients and seizure types.
 
Evaluation Metrics
Consider that your algorithm is developed to detect preictal periods that start 1h5min before seizures, and that a warning is useful if it is raised at least five minutes before seizures. This means that a seizure occurrence period (SOP) of 1h and an intervention time (IT) of 5 min are considered. 

Following metrics will be considered:

●    Sensitivity(%): Defined as the ratio of predicted seizures.
●    Time in warning (TiW) (%): Defined as the fraction of time spent in alarm.
●    Improvement over chance (IoC) (%)
●    Area Under the Receiver Operating Characteristic Curve (AUC-ROC).
●    False Alarm Rate (FAR) to assess the algorithm’s reliability.
Latency: Time difference between predicted pre-ictal periods and actual seizure onset.

Algorithms will be ranked using AUC-ROC

Competition Stages

●    Development Phase: Access to training data and annotated seizures onset times for model development. Data will be supplied as EDF files.
●    Submission of algorithms and models to the organizers. Submission details will be given by May 1st.

Publication policy

The competition organizers aim at a joint publication with the winning teams. We explicitly discourage use of the data for publications independent of this competition and this will not be covered by the download policies of the dataset. 

Incentive

●    Presentation of the algorithms of the teams at first and second position at the 5th International Congress on Mobile Health and Digital Technology in Epilepsy

●    Registration fee and hotel costs for the conference for one presenter of the two winning submissions. 

Who Should Participate?
 
This challenge is open to:

●    Data scientists with expertise in time-series analysis and Artificial Intelligence/Machine Learning.
●    Neuroscientists and clinicians interested in translational research.
●    Multidisciplinary teams combining domain knowledge and technical expertise.
 
Impact
 
The outcomes of this challenge have the potential to transform epilepsy management by enabling:

●    Early warnings for patients, improving their quality of life.
●    Development of closed-loop systems for seizure intervention.
●    Insights into the neurophysiology of seizure generation.

Timeline

●    Publication of the training dataset and reference algorithm: May 1st; 2025
●    Submission deadline for algorithms: June 30th, 2025
●    Running the algorithms on test dataset and ranking of contributions: July, 20th, 2025
●    Presenting of winner algorithms in person or online: 4 - 6 September 2025, at International Congress on Mobile Health and Digital Technology in Epilepsy
 
Join the Challenge
 
Be part of this groundbreaking effort to make seizures predictable and manageable. Collaborate, innovate, and push the boundaries of neurotechnology.
 
For more details on registration, dataset access, and timelines, visit https://epilongcontest.cisuc.uc.pt  or contact epilong@dei.uc.pt.