Deep learning in neuroimaging of epilepsy

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Deep learning (DL) offers advanced feature extraction from medical data, surpassing conventional machine learning.
• Neuroimaging modalities like MRI and PET are increasingly used in epilepsy research.
• Focal epilepsy, especially drug-refractory cases, benefits from advanced diagnostic and prognostic tools.

Purpose of the Study:

• To review the application of DL in neuroimaging for focal epilepsy, focusing on presurgical evaluation.
• To provide an overview of DL techniques and their use in epilepsy diagnosis, lateralization, and outcome prediction.
• To discuss current limitations, challenges, and future directions for DL in epilepsy research.

Main Methods:

• Review of existing literature on DL applications in neuroimaging for focal epilepsy.
• Theoretical overview of artificial neural networks and deep learning principles.
• Analysis of DL use in structural MRI (sMRI), functional MRI (fMRI), diffusion-weighted imaging (DWI), and PET.

Main Results:

• DL demonstrates utility in automated lesion detection, diagnosis, and lateralization of focal epilepsy.
• DL aids in presurgical evaluation and prediction of post-surgical outcomes in drug-refractory epilepsy.
• DL can analyze neuroimaging data considered negative by visual inspection, offering new insights.

Conclusions:

• DL in neuroimaging is a potential essential tool for individualized epilepsy treatment and understanding epilepsy as a network disorder.
• Wider multicenter collaboration and open-access tools are crucial for advancing DL in epilepsy research.
• DL can enhance clinical practice, particularly in challenging epilepsy cases and for predicting treatment success.