DeepEpiX: A software for visualization, annotation and automatic epileptical spike detection in MEG recordings

Area of Science:

• Neuroscience
• Medical Technology
• Artificial Intelligence

Background:

• Magnetoencephalography (MEG) is crucial for non-invasive brain activity recording, particularly in presurgical evaluation for drug-resistant epilepsy.
• Manual analysis of MEG data is labor-intensive and prone to variability, highlighting a need for efficient analytical tools.
• Accurate localization of the epileptogenic zone is essential for successful epilepsy surgery.

Purpose of the Study:

• To introduce DeepEpiX, a user-friendly software for automated analysis of MEG data.
• To facilitate the detection of pathological events, specifically Interictal Epileptogenic Discharges (IEDs), in epilepsy patients.
• To provide a platform for integrating and testing novel deep learning models for MEG analysis.

Main Methods:

• Development of DeepEpiX, an open-source software with a graphical interface for MEG data analysis.
• Integration of pre-trained deep learning models for automatic detection of Interictal Epileptogenic Discharges (IEDs).
• Implementation of a modular architecture to support the addition of new deep learning models and other data modalities.

Main Results:

• DeepEpiX achieved an F1-score of 47% for IED detection on data from the same acquisition system and 23% on data from a different system.
• Ensemble models demonstrated improved performance, increasing F1-scores by over 5% across both test sets.
• The software's performance indicates its capability in detecting IEDs, with potential for improvement through model ensembling.

Conclusions:

• DeepEpiX addresses the clinical need for efficient and automated MEG data analysis in epilepsy.
• The software's modular design allows for the integration and testing of new deep learning models.
• Future extensions could include support for other neuroimaging modalities like EEG and broader applications beyond epilepsy.