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

Magnetoencephalography (MEG) non-invasively records brain activity. It is widely used in presurgical evaluation of drug-resistant epilepsy, where it contributes to localizing the epileptogenic zone. Manual analysis of MEG is time-consuming and subject to inter-rater variability. Yet, there is a lack of adapted software facilitating its analysis.

We present DeepEpiX, a MEG data analysis software within a user-friendly graphical interface developed to assist clinicians in detecting pathological events, known as Interictal Epileptogenic Discharges (IEDs). The software allows applying basic preprocessing steps, while its main feature is the integration of pre-trained deep learning models for automatic IED detection. Its modular architecture supports the integration of novel models.

IED detection performances are reported on a use-case including 29 test patients with epilepsy. Model performance reaches a F1-score of 47% on test data, coming from the same acquisition system as the model training data, while the model is still capable of detecting IEDs on test data from a different acquisition system (F1-score: 23%). Furthermore, ensemble models improved F1-scores by more than 5% across both test sets, showing that different models capture complementary information about IEDs.

DeepEpiX is the only open source software integrating state-of-the-art deep learning methods with an intuitive graphical interface for automated MEG data analysis.

DeepEpiX answers clinical needs, while allowing the integration and test of novel deep learning models through the software. Although currently focused on MEG, its modular design could be extended to other modalities, such as EEG, and other applications.