Embedded System for Responsive Optogenetic Control of Spontaneous

Seizures in a Preclinical Temporal Lobe Epilepsy Model

Soﬁe Lasure

1,2 a

, Lien De Schaepmeester

2 b

, Sielke Caestecker

2 c

, Jeroen Spanoghe

2 d

Marijke Vergaelen

2 e

, Rik Verplancke

1 f

, Johannes Vierock

3 g

, Robrecht Raedt

2 h

and

Pieter Bauwens

1 i

Center for Microsystems Technology, Imec and Ghent University, Belgium

4BRAIN, Department of Head and Skin, Ghent University, Belgium

Neuroscience Research Center, Charit

e, Universit

atsmedizin zu Berlin, Germany

ﬁ

Keywords:

Epilepsy, Responsive Neuromodulation, Optogenetics, MSP430 Microcontroller, Embedded System.

Abstract:

Aims: Responsive neuromodulation employing optogenetics is a promising therapy which provides spatial and

temporal speciﬁcity for temporal lobe epilepsy (TLE), a neurological disorder characterised by the occurrence

of spontaneous seizures. In this study, we evaluated whether seizures could be detected with a self-designed

minimalist embedded system and terminated through activation of the WiChR opsin in the intrahippocampal

kainic acid (IHKA) mouse model of temporal lobe epilepsy. Methods: Mice were injected in the hippocam-

pus with kainic acid to simulate TLE and with the AAV2/9 viral vector to induce expression of the WiChR

opsin. Intracranial EEG was recorded and processed with a low-power microcontroller to detect the seizure

via the amplitude correlation metric. Upon detection, a 473 nm Light Emitting Diode (LED) was activated to

illuminate the hippocampus through an optrode. Results: It was possible to responsively illuminate seizures

with the embedded system and achieve a signiﬁcant reduction in seizure duration with a pulse train of 10 Hz,

5 ms, 10 mW for 90 s. A brief parameter study was performed although preliminary results were inconclusive.

Conclusions: In this study, we prove that we can responsively suppress seizures in the IHKA mouse model

within the limitations of a minimalist embedded system. Additionally, the WiChR opsin has been demon-

strated to have a high potential for efﬁcient seizure suppression with limited illumination.

1 INTRODUCTION

Temporal lobe epilepsy (TLE) is a chronic neuro-

logical condition characterised by recurrent seizures

originating in the temporal lobe of the brain. While

the use of anti-seizure medication remains the ﬁrst

line of treatment, neuromodulation offers a promis-

ing alternative for patients whose seizures are not ad-

equately controlled by medication or for whom re-

sective surgery is not an option. Neuromodulation

https://orcid.org/0000-0001-8625-5896

https://orcid.org/0009-0000-6660-0541

https://orcid.org/0000-0001-7398-183X

https://orcid.org/0000-0002-4427-0948

https://orcid.org/0000-0002-2915-667X

https://orcid.org/0000-0002-6642-9454

https://orcid.org/0000-0001-7368-5539

https://orcid.org/0000-0002-8939-0169

https://orcid.org/0000-0001-9398-2485

is a technique where the brain is stimulated, usually

electrically, and has been proven a viable therapy for

several neurological disorders, including TLE. Both

Deep Brain Stimulation (DBS) and the Responsive

Neuromodulation System (RNS) have been FDA-

approved and have shown promise in reducing seizure

frequency and improving quality of life for patients

(Ryvlin and Jehi, 2021; Ebrahim and Tungu, 2022).

However, there remain signiﬁcant drawbacks to

the use of electrical stimulation. The electrical pulses

affect a large area of the brain, including neurons

and neuronal projections not related to the neurolog-

ical disorder. Optical stimulation can provide a so-

lution as the stimulated area is much smaller, allow-

ing stimulation with an increased spatial speciﬁcity.

Optical stimulation would have a limited effect on

the brain activity so it must be used in combination

with techniques like optogenetics or optopharmacol-

ogy (Ebrahim and Tungu, 2022).

With optogenetics, light sensitive proteins

Lasure, S., De Schaepmeester, L., Caestecker, S., Spanoghe, J., Vergaelen, M., Verplancke, R., Vierock, J., Raedt, R. and Bauwens, P.

Embedded System for Responsive Optogenetic Control of Spontaneous Seizures in a Preclinical Temporal Lobe Epilepsy Model.

DOI: 10.5220/0013377000003911

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 18th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2025) - Volume 1, pages 88-99

ISBN: 978-989-758-731-3; ISSN: 2184-4305

(opsins) are introduced through the use of a viral

vector. These opsins can be activated by illumination,

either exciting or inhibiting a cell’s activity (Medina

et al., 2023). The WiChR opsin is a novel inhibitory

channelrhodopsin that introduces light-sensitive

potassium ion channels, maximally activated by light

pulses with a wavelength around 470 nm. Compared

to other opsins, the WiChR opsin has some major

advantages as the potassium ion channels have a

larger photocurrent during illumination and the

opsins close relatively slowly, allowing for efﬁcient

(Vierock et al., 2022).

Optopharmacology uses light-sensitive com-

pounds which can be locally activated with optical

stimulation. An example of this type of compound is

photocaged N6-cyclopentyladenosine (cCPA) which

can be activated by 405 nm illumination (Craey et al.,

2022; Spanoghe et al., 2024). A novel epilepsy treat-

ment is under development using optopharmacology

in combination with the use of evoked potentials

(EP’s), which are neuronal responses to a stimulation

pulse and are linked to the excitability of the brain.

Controlling the excitability, as extracted from the

EP’s, using responsive activation of cCPA has been

shown to reduce the occurrence of seizures ex vivo

(Craey, 2022).

The ﬁnal goal when developing responsive neu-

romodulation therapies is twofold: (1) a proof-of-

concept of the therapeutic effect and (2) the ability

to run these therapies on an implantable device. Both

goals present their own unique challenges but are in-

herently linked to each other through the required

hard- and software. The aim of the development of

this embedded system was focused towards an im-

plantable device but adjusted to allow neuromodula-

tion experiments within a preclinical setting. There-

fore, there were two main design considerations: (1)

implementation within an existing EEG recording set-

up and (2) the possibility for miniaturisation to allow

development towards an implantable device.

This paper will describe the design of hard- and

software of this embedded system. This device was

used to conduct a feasability study of a neuromodula-

tion therapy where seizures were suppressed upon de-

tection in the Intrahippocampal Kainic Acid (IHKA)

mouse model of TLE using the novel WiChR opsin,

illustrating that this is possible even with minimalist

hard- and software (Zeidler et al., 2018). Within this

experiment, an initial exploration of the stimulation

parameter space was performed to increase the un-

derstanding of the behaviour of the WiChR opsin and

search for the most efﬁcient illumination protocol.

2 METHODS

2.1 Embedded System

A major design consideration in the development of

this particular embedded system was the outlook for

development towards a thin, ﬂexible, implantable de-

vice. To facilitate this, the number of components had

to be minimised and the components had to be avail-

able in the smallest packaging factor, i.e. as bare dies

for the Integrated Circuits (IC’s) which was a major

limitation in component selection.

Additionally, to ensure a seamless interface within

the existing recording set-up and allow usage as a

more general purpose responsive neuromodulation

system, both hard- and software were designed with

ample connectors and possibilities to modify the sys-

tem depending on a given experiment. The embedded

system takes the ampliﬁed and ﬁltered EEG signals

as input and generates an output signal used to drive

a stimulation system. The embedded system gener-

ates output signals which are recorded together with

the EEG-signals to allow monitoring of the function-

ing of the responsive neuromodulation system and ob-

tain a synchronised recording of both. The integration

of the embedded system within the electrophysiology

set-up is illustrated in Figure 1A.

2.1.1 Hardware

The embedded system is based upon designs found

in literature and designed around a low-power micro-

controller unit (MCU) of the MSP430-family from

Texas Instruments (Pederson et al., 2019; Shon et al.,

2017). Besides the microcontroller and the required

components for the programming circuitry, there are

voltage regulators present which convert a single sup-

ply voltage into a double supply rail and generate

3.3 V as required by the MCU. Incoming signals pass

through a voltage offset circuit of 1.65 V to offset the

incoming signals and to ensure they are within the

Analog-to-Digital Converter (ADC) input range, as

shown in Figure 1B. On the output side, two Digital-

to-Analog Converters (DAC’s) are used. The ﬁrst is

used to transmit the metric used for the responsive

neuromodulation to use as feedback on how the em-

bedded system is performing. The second DAC is

used to control the stimulation parameters. Using a

switch, a choice can be made between voltage and

current pulses. Voltage pulses are generated using a

negative offset and an ampliﬁer as shown in Figure

1C. These can be used for direct stimulation or used

as a trigger for an external stimulation source, e.g. a

laser or Light Emitting Diode (LED) driver.

Embedded System for Responsive Optogenetic Control of Spontaneous Seizures in a Preclinical Temporal Lobe Epilepsy Model

Figure 1: A: Schematic overview of electrophysiology set-up and integration of embedded system within. Build-up of em-

bedded system and picture of realisation. NIDAQ: National Instruments Digital Acquisition Device. B: 1.65 V voltage offset

circuit. C: Voltage pulse generator circuit including -1.65 V offset. D: Howland current source including -1.65 V offset.

Current pulses are generated using a Howland cur-

rent source as shown in Figure 1D (Pederson et al.,

2019; Instruments, 2008). The circuit delivers an out-

put current as shown in Equation 1 where the second

factor is negligible if the R

= R

= k. Using

the resistor R

, the maximal stimulation current can

be dimensioned but a trade-off must be found between

the range of the output current and the resistance

of the stimulation electrode since they are related as

shown in Equation 2. For a stimulation electrode re-

sistance R

electrode

of 10 kΩ and an opamp supply rail

of 10 V, which results in an opamp output swing and

thus V

out

of 8 V, the output current will be limited to

800 µA. V

is limited to +/-1.65 V corresponding to

the output of the MCU DAC after the -1.65 V off-

set which means R

will be dimensioned at 825 Ω for

= R

= 200kΩ and R

= R

= 40kΩ. When di-

mensioning for increasing resistances of the stimula-

tion electrode, R

will have to be increased and the

maximal output current will decrease. Increasing the

voltage supply rails can mitigate this trade-off (Instru-

ments, 2008).

out

= V

out

(

−

)

(1)

with k =

out

= V

out

electrode

(2)

A pin header is used to connect an MSP MCU pro-

gramming and debugging probe (MSP-FET, Texas In-

struments, Texas, USA) to reprogram the MCU. Pin

headers, as well as BNC-connectors can be used to

connect incoming and outgoing signals

2.1.2 Seizure Detection Algorithm

Considering the computational limitations of the

MSP430F1611 microcontroller, a relatively simple

yet robust seizure detection algorithm is needed.

Seizures in the IHKA mouse model are characterised

by high frequency spiking activity observed in in-

tracranial EEG-signals, which can be detected using

peak detection. However, as this is computation-

ally expensive, an alternative metric called Amplitude

Correlation (AC) is used as described in literature

(Krook-Magnuson et al., 2013; White et al., 2006).

The amplitude correlation metric is based upon

the use of the signal range in a window with e.g. a

length of 120ms. For each window, the overlap of

the signal range is calculated with respect to those of

the previous N windows. In other words, the ampli-

tude correlation will only be high if the signal range

in the current window is high and similar to the range

in previous window(s), hence the term ”correlation”,

which is illustrated in Figure 2 for a window size of

1 s. As seizures are characterised by a prolonged pe-

riod (≥5 s) of high frequency spiking (Zeidler et al.,

BIODEVICES 2025 - 18th International Conference on Biomedical Electronics and Devices

Figure 2: The amplitude correlation (AC) is calculated as

the overlap of the signal range in current window S

with

the previous N windows, here illustrated for a window size

of 1 s.

2018), a summation of the amplitude correlation is in-

troduced to avoid a response to brief periods of spik-

ing activity which would be considered a false posi-

tive. High frequency spiking activity for a prolonged

period will increase the summed amplitude correla-

tion and based on a threshold crossing, a seizure de-

tection is initiated. Mathematically, this is expressed

as the following formulas with S representing a win-

dow:

Amplitude correlation AC(S

) = (3)

min[max(S

),max(max(S

n−1

),max(S

n−2

))]−

max[min(S

),min(min(S

n−1

),min(S

n−2

))]

summed,n

sum

∑

i=1

n−i

(4)

The amplitude correlation can be tuned by mod-

ifying (1) the window size, (2) the number of win-

dows to calculate the amplitude correlation with and

(3) the number of windows to consider in the summa-

tion. The window size and the number of windows

used to calculate the amplitude correlation affect the

minimal epileptic peak frequency needed for detec-

tion. With windows of 120 ms and 3 windows to con-

sider as used by Krook-Magnuson et al., the spiking

frequency has to be at least 5.5 Hz to have an impact

on the amplitude correlation as seen in Equation 5

(Krook-Magnuson et al., 2013). In the IHKA mouse

model, a minimum spiking frequency of 2 Hz is typi-

cally used to classify seizures (Zeidler et al., 2018) so

the amplitude correlation settings are modiﬁed to use

6 windows instead of 3 (including the current win-

dow). Modifying the number of windows included in

the summation affects the minimal required duration

of spiking activity and is set to 42 to correspond to a

period with spiking activity with a minimal duration

of 5 s. With these settings, a seizure with repetitive

spiking of 2 Hz will be detected within approximately

5 s and a seizure with a higher frequency will be de-

tected faster in about 2 s as more windows will have a

high AC.

Spiking f req. =

2 spikes

3 windows · 120ms

= 5.5Hz (5)

2.1.3 Software Implementation

The implementation of the previously described al-

gorithm is relatively simple but becomes increasingly

complex with the inclusion of various features to fa-

cilitate neuromodulation experiments as it must in-

clude the following options: (1) change the detec-

tion/stimulation parameters for each detection, (2) en-

able randomisation of sequence of various elements,

(3) disable stimulation after a set amount of time, (4)

re-enable stimulation with another set of elements and

(5) the option to conﬁgure settings during an ongoing

recording.

The incoming EEG-signals are recorded per win-

dow with a duration of 120ms, during which the cur-

rent minimum and maximum are stored. Upon ﬁnish-

ing the window, a ﬂag is set to initiate the calculation

of the amplitude correlation. If this is above the ”up”

threshold, an element is started, which can be either a

detection or a stimulation element. A stimulation ele-

ment consists of a pulse train with a certain amplitude,

pulse duration, pulse frequency, stimulation duration

as well as an additional ﬂag which can be used for

special settings, e.g. reduction of pulse duration after

X seconds. An element is only considered as ”ﬁn-

ished” when the amplitude correlation drops below a

”down” threshold to ensure a seizure is fully ﬁnished

prior to starting a new element. The elements are de-

ﬁned within a parameter set which speciﬁes the num-

ber of elements and their parameters. As the sequence

of the elements could have a confounding effect on

the results, the sequence is randomised before the ex-

ecution of each series of elements. Finally, there is the

option to customise parameters using a 12-position

switch. The MCU periodically checks the settings

of this switch and adjusts the parameters accordingly.

The exact changes are experiment-dependent and pre-

programmed prior to each experiment.

The software is written in C in Code Composer

Studio (Texas Instruments, Dallas, Texas, USA) and

programmed onto the microcontroller using the MSP-

FET debugging probe (Texas Instruments, Dallas,

Texas, USA). The structure of the software is based

around a simple main-loop and extensive use of in-

terrupts as most of the actions are time-dependent.

Two timers are used, both derived from the Mi-

crocontroller Unit (MCU) clock. The ﬁrst timer is

conﬁgured with a higher clock frequency as it will

control the timing of the ADC (sampling frequency

2kHz) and the stimulation pulses (switching speed

up to 1 kHz). The second timer uses a lower clock

frequency for the long term timing (range seconds-

minutes) to control long-term changes between pa-

rameter sets. An ADC interrupt will control digiti-

sation of the incoming signals. The maximum clock

Embedded System for Responsive Optogenetic Control of Spontaneous Seizures in a Preclinical Temporal Lobe Epilepsy Model

frequency of the MCU is 4.9 MHz so allows for pulses

down to 0.2 µs which is sufﬁcient for most stimula-

tion protocols as optical stimulation usually employs

pulses ≥0.5 ms and current pulses for evoked poten-

tials use segments of 200 µs.

2.2 Responsive Seizure Suppression in

the IHKA Mouse Model

2.2.1 Surgical Interventions

All experiments were conducted with male C57Bl/6J

mice (n=12, Janvier, The Netherlands) acquired at

an age of 7 weeks. Animals were treated according

to the European guidelines (directive 2010/63/EU).

All experiments described in the following were ap-

proved by the Animal Experimental Ethical Commit-

tee of Ghent University (ECD 22-39) and complied

with the ARRIVE guidelines. Mice were housed

under environmentally controlled conditions (humid-

ity: 40–60 %, temperature: 20–23 °C), a controlled

12 h/12 h light/dark cycle and with ad libitum access

to food and water.

For the injections and implantations, the animals

were secured in a stereotactic frame (Neurostar Robot

Stereotaxic system, Neurostar, Tubingen, Germany)

using ear bars and brought under anaesthesia, isoﬂu-

rane (induction 5 %, maintenance 2 %) and medi-

cal O2 (induction 1 l/min, maintenance 0.8 l/min).

Injection of 50 nl kainic acid (200 ng kainic acid

in 50 nl saline) was done to induce status epilepti-

cus with a ﬂow rate of 250 nl/min (Nanoliter 2020,

WPI, Florida, United States) at the stereotactic coor-

dinates AP-2.0 mm relative to bregma; ML+1.5 mm

relative to bregma; DV-1.8 mm relative to dura

(Desloovere et al., 2021). Three weeks thereafter,

1000 nl of the pAAV2/9.hsyn-WiChR1-ts-mScarlet-

er-WPREneu viral vector (titer 8.47E+12 VG/ml) was

injected with a ﬂow rate of 100 nl/min at the same

location. After 2 weeks, the optrode was implanted

at the same location with a depth of DV-1.8 mm rel-

ative to dura. The optrode consisted of a bipolar

depth electrode formed by twisting polyimide-coated

stainless steel wires (ø70 µm, California Fine Wire,

USA) and assembling these with an implantable op-

tical ﬁbre (ø400 µm, length 5 mm, 1.25 mm ferrule,

CFMLC14L05, Thorlabs, Germany) with the elec-

trode tips extending 300 µm beyond the tip of the

optical ﬁbre. An epidural screw electrode (stainless

steel, ø1.57 mm, Bilaney Consultants, Germany) was

placed contralaterally as ground/reference electrode.

After a week of recovery, the recording period was

started.

2.2.2 Electrophysiology

For EEG recording, the implanted electrodes were

referenced to stainless steel microscrew placed on

the contralateral side. The electrode leads and opti-

cal patch cord were connected to an electro-optical

swivel to allow free movement of the animals. These

signals were ﬁltered with a 2nd order high pass ﬁl-

ter at 0.159 Hz and ampliﬁed by a factor of 500 prior

to digitisation with a USB-6259 National Instruments

Data Acquisition device (NIDAQ) (National Instru-

ments, Austin, Texas, USA) which stored the ac-

quired data locally on a computer for ofﬂine analy-

sis. The digitisation had an accuracy of 16-bits over

a range of +/-10 V and thus has a resolution of +/-

305 µV, which means the resolution on the original

(non-ampliﬁed) signals was +/-305 µV / 500 = +/-

0.61 µV.

The outgoing signals from the embedded system

were digitised by the same USB-6259 National In-

struments data acquisition device, namely the ampli-

tude correlation used to detect the seizures and the

signal from a General Purpose Input Output (GPIO)-

pin, which corresponded to the detection/stimulation

action of the microcontroller as shown in Figure 1A.

In case of a seizure detection without stimulation, the

GPIO-pin was set to high during the detection. In case

of stimulation, the GPIO-pin was set to high/low in

accordance with the actual stimulation pulses that are

given.

The output of the responsive neuromodulation

system was a 5 V-pulse which connected to an LED

driver (T-Cube™LED driver, Thorlabs, Germany) via

a BNC cable, driving a 470 nm ﬁber-coupled LED

(M470F4 & M470F3, Thorlabs, Germany). This LED

was coupled to a rotary joint patch cord (ø400 µm,

NA 0.39, ceramic ferrule ø1.25 mm, RJPSL4, Thor-

labs, Germany), which coupled to the implanted optic

ﬁbre cannula with a ceramic sleeve (ADAL1, Thor-

labs, Germany). As relatively short pulses were used,

the accuracy had to be veriﬁed by measuring the op-

tical output power at the ferrule of the patch cord us-

ing a photometer (PM400, Thorlabs, Germany) prior

to implantation as shown in Figure 3. There was a

small delay at the on- and offset of the pulses but

the difference between the ideal and actual pulse was

small enough to be considered irrelevant for this ex-

periment.

2.2.3 Recording Protocol

The animals were attached to the recording set-up

non-stop for 6 days to minimise risks associated with

disconnecting and reconnecting the animals and the

associated risk of inter-day variability caused by a

BIODEVICES 2025 - 18th International Conference on Biomedical Electronics and Devices

Figure 3: Measurement of optical output power of stimula-

tion pulses of varying duration (0.5, 1, 2.5, 5) ms using T-

Cube LED driver and M470F4 LED (Thorlabs, Germany)

compared to target pulse.

varying coupling quality. Freedom of movement of

the animals was enabled by the use of the rotary patch

cord and electrical swivel.

In each stimulation day, a comparison between

a maximum of 6 parameters was made with the ad-

dition of 2 detection elements for reference. In the

IHKA mouse model, animals usually have a seizure

rate of 16 seizures/hour (Zeidler et al., 2018). During

an 8-hour recording, the occurrence of 160 seizures

could be expected which would result in 20 samples

of (non-)suppressed seizures for each stimulation el-

ement, providing sufﬁcient data to perform the re-

quired statistics. After 8 hours of responsive seizure

suppression, the system switches to seizure detection

only for 16 hours to negate potential long-term effects

of the illumination. The elements within a stimula-

tion day occurred in a randomised order to eliminate

the potential confounding effects of the stimulation

elements on each other. The elements were sampled

without repetition to ensure the same rate of occur-

rence for each element.

For the detection parameters, a detection thresh-

old for the amplitude correlation of 1.11 mV was cho-

sen as a baseline to provide a good balance between

detection speed and avoidance of false positives. Re-

ducing the detection threshold avoids false negatives

but provides difﬁculties with analysis as it would be

unclear whether the seizure was suppressed or there

would not have been a seizure in the ﬁrst place. The

”down” threshold to consider a seizure was ﬁnished

was set as the detection threshold minus 0.32 mV.

The starting point for the stimulation parameters

were based on those indicated in the original WiChR

publication by Vierock et al. (Vierock et al., 2022):

(1) a wavelength of 473 nm, (2) a pulse frequency

of 10 Hz and (3) a pulse duration of 5 ms. For the

stimulation power, 10 mW was chosen to ensure suf-

ﬁcient illumination of the hippocampus, as measured

at the ferrule of the rotary patch cord. After coupling

into the implanted ﬁbre, a loss of 11.6+/-4.1 % was

measured before implantation resulting in a mean out-

put power of 8.8 mW at the tip of the implanted ﬁbre

(not accounting for the inﬂuence of scar tissue forma-

tion due to the implantation). In combination with a

maximum duty cycle of 20 %, these settings provide

no concern for tissue heating as this would be lim-

ited to 1 °C according to simulations by Acharya et al

(Acharya et al., 2021). The stimulation duration was

ﬁxed at 90 s, corresponding to the 99 % percentile of

seizure duration seen in the IHKA mouse model to en-

sure the entire seizure was suppressed (Zeidler et al.,

2018).

The parameters of interest to vary during this

study were the pulse frequency and pulse duration as

indicated by early parameter explorations and single

neuron simulations by Weyn et al. which showed that

an increased pulse frequency increases the inihibitory

effect while maintaining the duty cycle (Weyn et al.,

2024). To study these parameters, a recording pro-

tocol was established as illustrated in Figure 4. The

ﬁrst day served to validate the autonomous seizure

detection. In the following 4 days, the pulse duration

was varied with each detection in the range 1, 2, 3,

4, 5 and 6 ms for four different detection thresholds:

1.11, 1.03, 0.95 and 0.87 mV as the speed of detec-

tion was hypothesised to affect the minimal effective

pulse duration. For day 2, the highest pulse duration

that did not work (T

suppression

) was extracted as the

lowest pulse duration resulting in a signiﬁcant reduc-

tion in seizure duration (Welch’s t-test, p<0.05) and

resulting in a seizure duration reduction of ≥75 %.

On day 6, both the pulse duration and frequency

were varied. The effective pulse duration T

suppression

was repeated along with T

suppression

/2 to validate the

(non)-effectiveness. Based on T

suppression

, the pulse

frequency was multiplied with a factor x2.5 and x5,

adjusting the pulse duration to maintain the duty cy-

cle. Finally, an element is tested where the stimu-

lation quantity is reduced after 5 s by decreasing the

pulse duration with a factor 1/2 to verify whether the

suppressive effect can be maintained with reduced

stimulation after the initial suppression. An example

of these variations with T

suppression

of 5 ms is illus-

trated in Figure 4.

2.2.4 Immunohistochemistry

Immunohistochemistry, a technique using antibod-

ies to visualise proteins in tissue, was performed to

analyse the hippocampal structure and the expres-

sion of the WiChR opsin. The protocol as previ-

ously described, (Desloovere et al., 2021), was fol-

lowed, with different antibodies to stain for microglia

(Iba-1) and RFP (Red Fluorescent Report protein), the

tag attached to the WiChR opsin. For Iba-1, rab-

Embedded System for Responsive Optogenetic Control of Spontaneous Seizures in a Preclinical Temporal Lobe Epilepsy Model

Figure 4: Schematic of recording protocol to evaluate inﬂuence of illumination parameters where each block represents a

(suppressed) seizure. Unless speciﬁed otherwise, illumination is done with a power of 10 mW, a pulse frequency of 10 Hz and

a stimulation duration of 90 s. The elements within a day are randomised within each repetition. After 8 hours of responsive

stimulation, the protocol switches to detection only for at least 16 hours.

bit anti Iba1 (Abcam Limited, Cambridge, United

Kingdom, 1:1000) was used as the primary anti-

body and goat anti-rabbit 488 nm (Novus Biologi-

cals, Abingdon, United Kingdom, 1:1000) as the sec-

ondary antibody. Staining for RFP was performed

using chicken anti-RFP (Thermo Fisher Scientiﬁc,

Massachusetts, United States, 1:1000) as the primary

antibody and goat anti-chicken 594 nm as the sec-

ondary antibody (Abcam Limited, Cambridge, United

Kingdom, 1:1000). Slides were scanned with a ﬂu-

orescent microscope (Panoramic 250, 3D Histech,

40×magniﬁcation) and automatically stitched.

2.2.5 Data Analysis

All data was manually annotated in a blind manner

(i.e. by only viewing the animal’s EEG signals with-

out the signals from the responsive neuromodulation

system). A seizure was quantiﬁed as repetitive spik-

ing with a minimum spiking frequency of 2 Hz and a

minimum duration of 7 s. A seizure was considered

ﬁnished if followed by a period of 7 s without any

spikes. Low frequency spiking activity which gen-

erally precedes a seizure was not annotated as part

of the seizure. To quantify the detection quality, the

manual and automatic annotations obtained from day

1 in the recording protocol (Figure 4) are compared

by calculating the percentage of correspondence, the

sensitivity and speciﬁcity.

Seizures with optogenetic inhibition were anno-

tated as described before if possible with the fol-

lowing considerations: (1) lower amplitude repetitive

spiking during the stimulation was considered to be

part of the same seizure as it is an indication that the

seizure is only partially suppressed and (2) high fre-

quency spiking activity during stimulation ≥5 s after

the original seizure ended, with duration ≥5 s were

annotated as a new seizure.

For each recording day, there were a number

of samples for each set of parameters within that

day. These were compared within an animal to ver-

ify whether they were signiﬁcantly different using

Welch’s t-test (p ≤0.05 to reject null-hypothesis) as

the variances proved unequal according to Levene’s

variance test (p≤0.05). To mitigate the inﬂuence of

the detection speed on the results, the time to detec-

tion was subtracted from the seizure duration. As a

consequence, the output metric would be more accu-

rately called the ”seizure duration after stimulation”

but the term ”seizure duration” will be used for con-

ciseness.

3 RESULTS

Of the initial 12 mice, only 2 showed the sponta-

neous seizures as expected in the IHKA mouse model,

coded SL038 and SL040, and could be used to anal-

yse the seizure detection and consequent optogenetic

seizure suppression. This sample size is too small for

inter-animal analysis and statistics so all following re-

sults remain limited to a comparison within each ani-

mal.

BIODEVICES 2025 - 18th International Conference on Biomedical Electronics and Devices

Figure 5: Comparison of seizure characteristics between

SL038 and SL040 with histograms of seizure duration and

a timeline displaying the occurrence of seizures across the

10 hour recording.

3.1 Seizure Detection

Firstly, a comparison was made between the online

calculation of the amplitude correlation by the MCU

and an ofﬂine calculation performed in Python using

the same settings for the computation to validate the

online calculation by the embedded system. For a

1-hour recording, this showed a correlation between

the online and ofﬂine correlation with a Pearson R²

of 97.5 % and a mean error of 0.82 µV, illustrating a

strong correspondance between the online and ofﬂine

calculations.

Table 1 shows the results of the autonomous

seizure detection of these two animals. A clear dif-

ference can be seen between the two animals under

consideration, SL038 and SL040, which can be at-

tributed to a difference observed between the charac-

teristics of the spontaneous seizures as illustrated in

Figure 5. SL038 shows a high number of seizures

>20 seizures/hour with a relatively short duration.

Figure 7A shows a representative EEG-trace with

these repetitive occurrences of high frequency spiking

activity which fall just around the criteria to be classi-

ﬁed as seizures. The seizure detection algorithm was

tuned to respond to activity longer than 5 s so encoun-

ters difﬁculties annotating these short seizures with a

duration between 5-10 s, leading to a relatively low

sensitivity of 42.5 %. Despite this, the seizure detec-

tion algorithm achieved an accuracy of 91.3 % and a

speciﬁcity of 98.0 %.

SL040 displays longer and less frequency sponta-

neous seizures with an average duration 36.5 s and a

rate of almost 10 seizures/hour (Zeidler et al., 2018).

These were easier to detect automatically, which is re-

ﬂected in the increased accuracy (97.8 %), sensitivity

(85.8 %) and speciﬁcity (99.0 %) compared to SL038.

For both animals, the optimisation towards the avoid-

ance of false positives is reﬂected in a higher speci-

ﬁcity compared to the sensitivity.

Table 1: Results of autonomous seizure detection using the

amplitude correlation metric and a detection threshold of

1.11 mV.

SL038 SL040

Nr of seizures 216 97

Mean seizure duration 19.3 s 36.5 s

Standard deviation +/-11.6 s +/-13.2 s

Accuracy 91.3 % 97.8 %

Sensitivity 42.5 % 85.8 %

Speciﬁcity 98.0 % 99.0 %

Median time to detection 5.4 s 4.0 s

3.2 Seizure Suppression

Figure 7A-B shows EEG traces for SL038 and SL040

showing representative examples of spontaneously

occurring and suppressed seizures. For SL038, an

immediate suppression can be seen without the occur-

rence of spiking activity during the stimulation period

for a pulse duration of 5ms. For SL040, the effect

of the illumination does not lead to complete seizure

suppression as shown by a comparison of the resulting

seizure duration in Figure 6. The reduction in seizure

duration for 5ms pulses is 51.6 %, which is below the

target of 75 %. Therefore, the protocol in Figure 4

was modiﬁed for SL040 to use a pulse duration of 2.5,

5, 7.5, 10, 15 and 20 ms instead of the original 1, 2, 3,

4, 5 and 6 ms. The difference in effect between SL038

and SL040 could be linked to the differences seen in

the characteristics of their spontaneous seizurs.

Figure 7C-D shows a detailed image of the ef-

fect of the illumination on the seizure duration, illus-

Figure 6: Comparison of resulting seizure duration for

a varying stimulation pulse duration between SL038 and

SL040.

Embedded System for Responsive Optogenetic Control of Spontaneous Seizures in a Preclinical Temporal Lobe Epilepsy Model

Figure 7: A-B: Typical seizures and effect of 10 Hz illumination with insufﬁcient and sufﬁcient pulse duration for SL038 and

SL040. C-D: Histograms of resulting seizure duration for a varying pulse duration for SL038 and SL040. E-F: Box plot of

resulting seizure duration for various detection thresholds for SL038 and SL040.

BIODEVICES 2025 - 18th International Conference on Biomedical Electronics and Devices

Figure 8: Box plot of reduction in seizure duration for

various combinations of pulse duration and frequency for

SL040.

trating the inter seizure variability, especially for the

shorter pulses and especially for SL040. For SL038,

every pulse duration ≥2ms has a signiﬁcant effect

(p≤0.0001) and pulses ≥5 ms result in a mean result-

ing seizure duration of 3.5 s, or a reduction of 85.5 %

compared to no stimulation which indicates a com-

plete, immediate suppression and was hence selected

as the T

suppression

pulse duration. The same analy-

sis is performed for SL040 and shows a signiﬁcant

difference between the unstimulated and stimulated

seizures (p≤0.0001) for each pulse duration. How-

ever, only a pulse duration ≥15 ms achieves a reduc-

tion in seizure duration ≥70 % and therefore, 20 ms

was chosen as T

suppression

pulse duration to ensure suf-

ﬁcient suppression.

A comparison of the stimulation effect between

the various thresholds can be seen in Figure 7E-F and

shows no signiﬁcant difference in the effect on the

seizure between the varying thresholds (p>0.05). A

threshold of 1.11 mV is not included in the analysis

for SL040 as only 14 seizures occurred during these

8 hours of responsive neuromodulation, resulting in

insufﬁcient data quantity for an accurate comparison.

The analysis of stimulation day 6 with varying

pulse duration and frequency could only be performed

for SL040 due to a loss of signal in SL038, likely

cause by a displacement of the depth electrode. For

SL040 there is a signiﬁcant difference between un-

stimulated and stimulated seizures (p≤0.05) but no

signiﬁcant difference between the various stimulation

settings (p>0.05) as illustrated in Figure 8. This re-

sult is questionable however as these results were only

obtained for a single animal, which showed a reduced

stimulation effect compared to SL038 so this must be

considered as an initial observation. Additional data

is required from multiple animals to attain a decisive

conclusion, especially considering the evident inter-

animal variability in seizure characteristics.

3.3 Immunohistochemistry

Of the original 12 mice, only 2 showed spontaneous

seizures which is below expectation as the IHKA

mouse model is generally very reliable (Zeidler et al.,

2018). An explanation had to be found to ensure

the safety of the embedded system and this could

be found in the results of the immunohistochem-

istry. Initial analysis of these images shows that from

the 10 animals without spontaneous seizures, 4 have

either no hippocampal lesion or show very limited

degradation of the neuronal cell layers in the dentate

gyrus, illustrated in Figure 9A, which explains the

lack of spontaneous seizures. The remaining 6 ani-

mals showed a severe hippocampal lesion where the

injected side was almost entirely degraded as shown

in Figure 9B, again providing a different but suitable

explanation for the lack of spontaneous seizures. Fig-

ure 9C shows the histology of SL040 exhibiting a

hippocampal lesion as expected in the IHKA mouse

model. None of the histology images showed any in-

dications there might have been an issue with the em-

bedded system, which could be seen as e.g. a corti-

cal lesion caused by an unwanted electrical pulse sent

through the electrodes

The inter-animal variability within this experi-

ment could be attributed to a variety of explanations

(or a combination thereof) which will be investigated

further, including (1) a potential issue with the injec-

Figure 9: Immunohistochemistry images of DAPI staining

in transversal plane around injection/implantation location.

A: Example of animal without seizures with minimal hip-

pocampal lesion. B: Example of animal without seizures

with major hippocampal damage. C: Example of animal

with seizures and typical kainic acid lesion.

Embedded System for Responsive Optogenetic Control of Spontaneous Seizures in a Preclinical Temporal Lobe Epilepsy Model

Figure 10: A: Schematic of biphasic current pulse. B: EEG-traces of evoked potentials averaged over 5 repetitions for

stimulation pulses generated by the DS4 stimulator (400 µA) and the embedded system (400 and 1200 µA).

tion of the kainic acid and/or the WiChR viral vector,

(2) potential toxicity due to the combination of the

WiChR viral vector and the kainic acid, (3) the rela-

tively high titer of the WiChR viral vector and/or (4)

a severe immune response triggered by the implanta-

tion of the relatively large optical ﬁbre (ø400µm). A

follow-up study will aim to tackle these potential is-

sues.

3.4 Evoked Potentials

A ﬁnal experiment was conducted with the aim to test

the ability of the embedded system to deliver accu-

rate biphasic current pulses to generate evoked poten-

tials for potential use in a responsive neuromodulation

experiment based upon brain excitability. The experi-

ment was performed simultaneously with an acute ex-

periment under anaesthesia as described by Spanoghe

et al. (Spanoghe et al., 2024). For stimulation, a

biphasic current pulse with a phase duration of 200 µs

and a current amplitude of +/-400 µA was generated

using the DS4 current stimulator (Digitimer, United

Kingdom) as shown in Figure 10A. At the end of the

experiment, the DS4 stimulator was replaced by the

embedded system to allow a direct comparison. Pre-

vious bench top testing showed that both systems pro-

vided an identical stimulation pulse over a 10 kΩ re-

sistor.

However, in vivo, there was a difference as the ad-

dition of the embedded system in the recording set-

up introduced noise within the recording, likely due

to grounding issues. These experiments were per-

formed under anaesthesia, which forms a less con-

trolled environment for EEG recording compared to

the environment for housing and chronic EEG record-

ings as used in the previously described chronic re-

sponsive neuromodulation experiment. As a conse-

quence, the 400 µA pulses from the embedded system

were more noisy and resulted in a different evoked

potential compared to the DS4 stimulator as shown in

Figure 10B for 5 averaged repetitions of this stimula-

tion pulse. The evoked potentials from the embedded

system show an increased stimulation artefact and the

evoked potential does not contain a population spike,

i.e. the negative peak generated by collective gener-

ation of action potentials by the neurons. When in-

creasing the stimulation amplitude to 1200 µA, it was

possible to generate an evoked potential with a popu-

lation spike.

4 CONCLUSION

The embedded system has been demonstrated to be

functional and can be used for responsive neuromod-

ulation experiments using electrical or optical stimu-

lation modalities though more optimisation is needed

for the use of biphasic current pulses with a short

phase duration. The design of this embedded system

shows the feasibility of performing chronic respon-

sive neuromodulation reliably with a minimalist hard-

and software design.

The seizure detection algorithm using the ampli-

tude correlation metric is simple yet surprisingly ef-

fective. An improvement can be made by the imple-

mentation of the use of an adaptive threshold based on

the use of a long-term average of the amplitude cor-

relation to mitigate inter-animal variability in signal

strength. For the future clinical application however,

this type of responsive neuromodulation using seizure

detection will be insufﬁcient as you allow the seizure

to start before intervening. Ideally, the seizure would

be predicted or forecasted and preventive stimulation

would be used (Mormann et al., 2007; Baud et al.,

2022).

The therapeutic potential of seizure suppression

using the WiChR opsin is evident as seen by the

quick suppressive effect using only limited stimula-

tion. However, experimental difﬁculties limited the

scope of this study to an analysis of only 2 animals

which is insufﬁcient to draw conclusions, especially

considering the inter-animal variability which was ev-

BIODEVICES 2025 - 18th International Conference on Biomedical Electronics and Devices

ident in this study. Future experiments will aim to in-

crease the number of animals included in the study

to allow more thorough analysis and statistics on the

seizure suppression and the inﬂuence of the stimula-

tion parameters.

AUTHOR CONTRIBUTIONS

SL - System design, data analysis, writing; LDS, JS,

MV, SC - Surgical interventions, animal welfare; JV

- Design and supply of WiChR viral vector; RV, RR,

PB - Supervision, funding acquisition

REFERENCES

Acharya, A. R., Vandekerckhove, B., Larsen, L. E., Del-

beke, J., Wadman, W. J., Vonck, K., Carette, E.,

Meurs, A., Vanﬂeteren, J., Boon, P., Missinne, J., and

Raedt, R. (2021). In vivo blue light illumination for

optogenetic inhibition: effect on local temperature and

excitability of the rat hippocampus. Journal of neural

engineering, 18.

Baud, M. O., Proix, T., Gregg, N. M., Brinkmann, B. H.,

Nurse, E. S., Cook, M. J., and Karoly, P. J. (2022).

Seizure forecasting: Bifurcations in the long and

winding road. Epilepsia.

Craey, E. (2022). A-novel-photocaged-adenosine-a1-

receptor-agonist-for-closed-loop-optical-control-of-

hippocampal-neurotransmission-and-suppression-of-

epileptiform-bursts-in-an-ex-vivo-slice-model-for-

epilepsy.

Craey, E., Hulpia, F., Spanoghe, J., Manzella, S., Larsen,

L. E., Sprengers, M., Bundel, D. D., Smolders, I.,

Carrette, E., Delbeke, J., Vonck, K., Boon, P., Calen-

bergh, S. V., Wadman, W. J., and Raedt, R. (2022).

Ex vivo feedback control of neurotransmission using

a photocaged adenosine a1 receptor agonist. Interna-

tional journal of molecular sciences, 23.

Desloovere, J., Boon, P., Larsen, L. E., Goossens, M. G.,

Delbeke, J., Carrette, E., Wadman, W., Vonck, K., and

Raedt, R. (2021). Chemogenetic seizure control with

clozapine and the novel ligand jhu37160 outperforms

the effects of levetiracetam in the intrahippocampal

kainic acid mouse model. Neurotherapeutics, 19:342.

Ebrahim, A. A. and Tungu, A. (2022). Neuromodulation

for temporal lobe epilepsy: a scoping review. Acta

Epileptologica, 4:1–11.

Instruments, T. T. (2008). An-1515 a comprehensive study

of the howland current pump (rev. a) — enhanced

reader.

Krook-Magnuson, E., Armstrong, C., Oijala, M., and

Soltesz, I. (2013). On-demand optogenetic control of

spontaneous seizures in temporal lobe epilepsy. Na-

ture Communications 2013 4:1, 4:1–8.

Medina, R., Ho, A., Reddy, R., Chen, J., and Castellanos, J.

(2023). Narrative review of current neuromodulation

modalities for spinal cord injury. Frontiers in Pain

Research, 4.

Mormann, F., Andrzejak, R. G., Elger, C. E., and Lehnertz,

K. (2007). Seizure prediction: the long and winding

road. Brain : a journal of neurology, 130:314–333.

Pederson, D. J., Quinkert, C. J., Arafat, M. A., Somann,

J. P., Williams, J. D., Bercich, R. A., Wang, Z., Albors,

G. O., Jefferys, J. G., and Irazoqui, P. P. (2019). The

bionode. ACM Transactions on Embedded Computing

Systems (TECS), 18:400.

Ryvlin, P. and Jehi, L. E. (2021). Neuromodulation for re-

fractory epilepsy. Epilepsy currents, 22:11–17.

Shon, A., Chu, J. U., Jung, J., Kim, H., and Youn, I. (2017).

An implantable wireless neural interface system for

simultaneous recording and stimulation of peripheral

nerve with a single cuff electrode. Sensors (Basel,

Switzerland), 18.

Spanoghe, J., Boon, P., Vergaelen, M., Colvenaer, M. D.,

Mariman, T., Vonck, K., Carrette, E., Wadman, W.,

Craey, E., Larsen, L. E., Sprengers, M., Missinne, J.,

Calenbergh, S. V., Bundel, D. D., Smolders, I., and

Raedt, R. (2024). Photopharmacological activation of

adenosine a1 receptor signaling suppresses seizures in

a mouse model for temporal lobe epilepsy. bioRxiv,

page 2024.07.04.602052.

Vierock, J., Peter, E., Grimm, C., Rozenberg, A., Chen,

I. W., Tillert, L., Scalise, A. G. C., Casini, M., Au-

gustin, S., Tanese, D., Forget, B. C., Peyronnet, R.,

Schneider-Warme, F., Emiliani, V., B

a, O., and

Hegemann, P. (2022). Wichr, a highly potassium-

selective channelrhodopsin for low-light one- and

two-photon inhibition of excitable cells. Science Ad-

vances, 8.

Weyn, L., Tarnaud, T., Schoeters, R., Becker, X. D., Joseph,

W., Raedt, R., and Tanghe, E. (2024). Computa-

tional analysis of optogenetic inhibition of ca1 neu-

rons using a data-efﬁcient and interpretable potassium

and chloride conducting opsin model. bioRxiv, page

2024.10.17.618665.

White, A. M., Williams, P. A., Ferraro, D. J., Clark, S.,

Kadam, S. D., Dudek, F. E., and Staley, K. J. (2006).

Efﬁcient unsupervised algorithms for the detection of

seizures in continuous eeg recordings from rats af-

ter brain injury. Journal of Neuroscience Methods,

152:255–266.

Zeidler, Z., Brandt-Fontaine, M., Leintz, C., Krook-

Magnuson, C., Netoff, T., and Krook-Magnuson, E.

(2018). Targeting the mouse ventral hippocampus in

the intrahippocampal kainic acid model of temporal

lobe epilepsy. eNeuro, 5:ENEURO.0158–18.2018.

Embedded System for Responsive Optogenetic Control of Spontaneous Seizures in a Preclinical Temporal Lobe Epilepsy Model