Files
Degand_75222100_2026.pdf
Open access - Adobe PDF
- 47.94 MB
Details
- Supervisors
- Faculty
- Degree label
- Abstract
- Epilepsy is a chronic neurological disorder affecting more than 50 million people worldwide, characterized by recurrent seizures resulting from abnormal brain electrical activity. For drug-resistant patients, surgical treatment may be considered, requiring precise localization of the epileptogenic zone. In this context, intracranial electroencephalography (iEEG) provides direct, high-fidelity recordings of brain activity and constitutes a key tool for both clinical decision-making and neuroscientific research. However, the preprocessing of iEEG data remains a challenging and poorly standardized task, typically requiring significant domain expertise and manual intervention. This thesis presents a semi-automated pipeline for location-aware iEEG signal preprocessing in epileptic patients, integrating two prior works developed at UCLouvain. The proposed pipeline covers the full preprocessing workflow: data preprocessing, electrode identification, electrode localization using the ElectroLoc algorithm developed Q. De Laet, anatomical contact labeling via an atlas, and epileptic seizure detection using a transformer-based model (MViT) implemented by T. Bary . A shift correction mechanism is introduced to handle mislocalized electrodes, and atlas dilatation is applied to improve anatomical labeling accuracy. The pipeline is validated on data from two patients implanted with iEEG depth electrodes. Results demonstrate competitive performance across all modules, with particular strengths in electrode localization and contact labeling. Limitations related to dataset size, shift correction conditions, and seizure detection generalization are discussed, along with perspectives for future improvements including few-shot learning approaches for seizure detection. The code produced for this thesis can be found in the following repository: https://github.com/maedegand/MasterThesis.git