SEIZURE

What SEIZURE does

Project overview

Fusing structural, metabolic and electrophysiological brain data with machine learning to find the epileptogenic zone and predict surgical outcome.

SEIZURE work package diagram. A shared multimodal database feeds three localization work packages, MRI and PET, MEG, and multimodal fusion, toward epileptogenic zone localization and surgical outcome prediction.
The SEIZURE analysis framework, from the shared multimodal database to epileptogenic zone localization and outcome prediction.

Work packages

WP0

Project management

Lead: C. Lartizien, J. Jung

Coordinate the consortium and the shared multimodal database.

WP1

Database collection

Lead: J. Jung (HCL)

Assemble a curated multimodal cohort of more than 200 patients across MRI, PET, MEG and clinical data.

WP2

EZ localization with MRI and PET

Lead: C. Lartizien (CREATIS)

Detect and phenotype subtle epileptogenic lesions from structural and metabolic imaging.

WP3

EZ localization with MEG

Lead: R. Quentin (CRNL), P. Borgnat (ENS de Lyon)

Detect interictal spikes and map spike propagation with graph signal processing.

WP4

Multimodal fusion and outcome prediction

Lead: C. Lartizien (CREATIS)

Fuse the modalities to localize the EZ and predict surgical outcome.

The challenge

Pinpointing the epileptogenic zone

Epilepsy resists drug treatment in roughly 30% of patients. For them, surgically removing the epileptogenic zone (EZ) is the best option, yet no single biomarker pinpoints it and predicting surgical outcome stays hard.

MRI and PET expose subtle structural and metabolic lesions, while MEG records the brief electrophysiological events that travel through the epileptic network. These anatomical and functional views have largely evolved in parallel, and epileptologists still combine them by hand.

MRI (structural), PET (metabolic) and MEG (electrophysiological) images converging through a neural network toward the epileptogenic zone shown on a brain.

Our approach

Statistical machine learning across modalities

SEIZURE builds a statistical analysis framework that fuses multiparametric MRI, PET, MEG and clinical data to localize the EZ and predict surgical outcome. It draws on advanced machine learning for image and graph signal processing and for combining heterogeneous data.

Our hypothesis is that this integrated, transdisciplinary approach can deliver a step change in diagnostic performance, built on a multimodal database of more than 200 patients already collected by the clinical partner.

200+
patients
3
imaging modalities
5
work packages
48
months, to Nov 2028

Timeline

2024
Project start and database collection
2025
EZ localization with MRI, PET and MEG
2026
Multimodal fusion models
2027
Outcome prediction and validation
2028
Clinical translation

Partners

A transdisciplinary consortium across imaging, neuroscience, clinical epileptology and signal processing.

ANR CREATIS CRNL ENS de Lyon HCL CNRS Inserm INSA Lyon Inria