Objective:

The objective of this study is to determine if a machine learning model can utilize magnetic resonance imaging (MRI) to differentiate patients with functional seizure disorder (FSD) from those with temporal lobe epilepsy (TLE).

Background:

FSD is a diagnosis associated with significant disability, but that is ultimately poorly understood. Although some studies have suggested anatomic commonalities among FSD patients, consistent MRI markers and underlying anatomic abnormalities have yet to be elucidated. Clear demonstration of these factors would result in easier diagnosis and may even aid in developing novel treatments for the condition. As the literature has yet to conclusively identify MRI markers, a machine learning model trained through MRI segmentation offers a novel method to classify radiographic markers that would not be detectable through traditional analysis methods.

Design/Methods:

After institutional review board approval, we retrospectively identified 64 patients with FSD and 23 with TLE who underwent workup in the epilepsy monitoring unit between 2015 and 2022. Patients were included in the study only if their typical episode was observed during EEG monitoring, thus confirming their diagnosis and study group. MRI for each patient was then pre-processed and systematically fed through a convolution neural network, generating a novel machine-learning model.

Results:

Although analysis and patient selection are still ongoing, our model has already identified several regions of interest. Currently, we have obtained a positive model certainty of 55% and a negative model certainty of 65% when training from the left thalamus, anterior corpus callosum, and left cerebral cortex.

Conclusions:

Our findings demonstrate promise in using machine learning when assessing complex disorders such as functional seizure disorder. Furthermore, continued work with our machine-learning model may aid in identifying anatomic MRI markers shared by patients with FSD. This work may ultimately expand the understanding of the underlying mechanisms and diagnostic methods available for the disorder.