Epilepsy Identification Using Biomarkers based on Visually Evoked EEG Responses.

Epilepsy is a chronic neurological disorder, characterized by recurrent
seizures and abnormal electrical brain activity, affecting approximately 1% of
the global population [Fiest et al. 2017, Beghi et al. 2019].
Electroencephalography (EEG) is an essential tool for recording brain activity
and can be used to detect changes that precede a seizure, indicating the
critical transition to a seizure state [Lopes Da Silva et al. 2003, Kramer et
al. 2012, Jirsa et al. 2014]. EEG biomarkers can help to estimate the
likelihood of an impending seizure, though the mechanisms underlying this
transition are still only partially understood [Meisel et al. 2015, Maturana et
al. 2020]. One current theory suggests that a disrupted balance between
excitation and inhibition among neurons leads to increased cortical
excitability, potentially resulting in hypersynchronous neuronal activity and
triggering seizures [Rogawski et al. 2008, Mantegazza et al. 2018, Bauer et al.
2021]. As the brain approaches this state transition, it becomes more sensitive
to perturbations and may take longer to return to equilibrium after such
disruptions [Scheffer et al. 2009]. Visual stimulation is used clinically as a
safe perturbation, particularly in individuals with genetically generalized
epilepsy, who may demonstrate a photoparoxysmal response (PPR) that can support
an epilepsy diagnosis [Fisher et al. 2022].

Previous studies, such as those by Kalitzin et al. (2002), Parra et al. (2003),
and Vranic-Peters et al. (2023), have investigated EEG biomarkers during visual
stimulation. These studies showed that photosensitive individuals (who exhibit
a PPR or seizure in response to visual stimulation) display increased
excitability, with an associated higher likelihood of PPR and seizures
[Kalitzin et al. 2002, Parra et al. 2003, Vranic-Peters et al. 2023]. By
reproducing and expanding upon these studies, we aim to confirm and enhance the
effectiveness of EEG biomarkers in detecting epilepsy through heightened
excitability during visual stimulation. Such biomarkers may contribute to
faster and more reliable epilepsy diagnoses, potentially reducing the need for
repeated EEG measurements, shortening hospital stays, and decreasing the time
needed to establish the correct treatment.

Primary Objective:
The primary objective to create a machine learning model based on biomarkers
analysed in EEG-responses during various types of visual stimulation that could
aid in the identification of EEGs of people with epilepsy and healthy
participants. We will focus on biomarkers previously tested during visual
stimulation by Kalitzin et al. (2002), Parra et al. (2003), and Vranic-Peters
et al. (2023), as well as on the biomarkers tested by Thangavel et al. (2022)
during the interictal phase.

Secondary Objectives:
To create a machine learning model based on these features that can distinguish
EEGs of people with focal and generalised epilepsy.
To create a machine learning model based on these features that can distinguish
EEGs of people with PNEA and healthy individuals.
To identify which features are the most important factors in the machine
learning model distinguishing epilepsy from heathy.

The study will last between a year to 1.5 years, to recruit a sufficient number
of participants with epilepsy at the Epilepsy Monitoring Unit (EMU) of SEIN
Heemstede.

EEG Recording:
At the EMU, the brain activity of participants is monitored to assist in the
diagnosis and evaluation of epilepsy. Throughout their stay, they are
continuously monitored via video surveillance and real-time EEG tracking. At
the end of their registration, before the EEG is disconnected, the measurements
for this study will be conducted. Healthy participants will be asked to come in
for a one-time EEG recording lasting about an hour.

The EEG will be recorded during a wide range of visual stimulation using three
different modalities: a stroboscopic flash lamp with flashes between 1 and 60
Hz, as standard in the clinic; LED goggles with chirp stimulation; and a screen
with alternating red-blue stimulation with varying modulation depth and
vertical stripes, with and without color.

EEG Analysis:
The recorded EEGs will be pre-processed according to a standardized procedure,
and biomarkers will be applied to the signals. The biomarker values are used as
parameters in a machine learning model. By splitting the group into a training,
test, and validation set, a proof-of-concept will be conducted to assess how
well the model performs in classifying EEGs of individuals with epilepsy and
healthy individuals. Similar appraches will be used to create a model that can
identify EEGs from people with PNEA and healthy individuals and one taht can
identify EEGs from poeple with focal and generalised epilepsy.

The intervention consists of the visual stimulation during the EEG recording.
This consists of three different modalities: a stroboscopic flash lamp with
flashes between 1 and 60 Hz, as standard in the clinic; LED goggles with chirp
stimulation; and a screen with alternating red-blue stimulation with varying
modulation depth and vertical stripes, with and without color.
This will last about half an hour. This is no therapeutic intervention, only
the direct response to the stimulation in the EEG will be evauated and
compared.

The burden for participants from the EMU is minimal. While participation may
slightly extend their admission time, the EEG electrodes are already in place,
and the patients are already on-site, so joining the study does not require
significant additional steps. Participants will not directly benefit from this
study.
For healthy participants, the burden is slightly greater as they need to travel
to the clinic and the EEG electrodes must be applied. This makes participation
somewhat more demanding for this group.
The main risk is the potential induction of seizures in response to visual
stimuli, although this risk is very low in healthy participants. It is
estimated that 2.2% of individuals with epilepsy are photosensitive
(light-sensitive) [Rathore et al. 2020]. However, not all people in this group
will experience a seizure; some may only exhibit a photoparoxysmal response
(PPR). If a generalised PPR is observed (which may be a precursor to a
seizure), the measurement will be stopped immediately.