



























45 J Global Clinical Engineering Vol.6 Special Issue 6: 2024

Conference Paper 

Novel Functional Electrical Stimulation Parameter Optimization 
for Neurorehabilitation Using Both Conventional and AI 
Techniques 

Arsenios Arsenidis1, Alexandros Moraitopoulos2, Alkinoos Athanasiou2, Alexandros Vildiridis3, Panagiotis Bamidis2, Petros 
Stefaneas4 and Alexandros Astaras5 

1Department of Physics School of Applied Mathematical and Physical Sciences, NTUA, Greece. 
2Lab of Medical Physics & Digital Innovation, AUTH, Greece. 
3Department of Business Administration, University of Piraeus, Greece. 
4Department of Mathematics School of Applied Mathematical and Physical Sciences, NTUA, Greece. 
5Computer Science Division of Science & Technology American College of Thessaloniki, Greece. 

* Corresponding Author Email: sl_arsenis@hotmail.com 

ABSTRACT

Neurological conditions such as stroke or spinal cord trauma often attenuate or disrupt nerve connections, leading to loss of 
muscle function, sensation, or responsiveness. The application of physical and occupational therapy rehabilitation protocols can 
help regain some of the lost functions and significantly improve a patient’s quality of life. These protocols leverage the principle 
of neuroplasticity, an inherent property of the brain that allows the formation of new neural connections in response to external 
stimuli. Electrical Muscle Stimulation (EMS) has been proven to amplify the effects of rehabilitation as it adds new stimuli in the 
form of suitable electric pulse-trains directly to the neuromuscular system. Certain rehabilitation protocols incorporate functional 
exercises that mimic natural movements, which can in turn benefit from the application of synchronized electric pulses. This 
process, known as Functional Electrical Stimulation (FES), has been demonstrated to be beneficial with respect to the nature and 
longevity of neuromuscular adaptations as well as brain reorganization. This paper considers techniques for the optimization 
of these parameters and presents preliminary in vivo experimental results demonstrating the proposed methodology. 

Keywords—Medical devices, Denervation, Stroke, Spinal cord injury (SCI), Functional Electrical Stimulation (FES), Func-
tional Electrical Stimulation Therapy (FEST), Medical instrumentation, Neurorehabilitation, Physical rehabilitation, Machine 
learning, Artificial intelligence (AI), Biomedical engineering, Central Nervous System (CNS), Peripheral Nervous System (PNS).

Copyright © 2024. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY): Creative Commons - 
Attribution 4.0 International - CC BY 4.0. The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright 
owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduc-
tion is permitted which does not comply with these terms.

http://www.globalce.org
http://globalce.org
http://globalce.org
mailto:sl_arsenis@hotmail.com
https://creativecommons.org/licenses/by/4.0/
https://creativecommons.org/licenses/by/4.0/


J Global Clinical Engineering Vol.6 Special Issue 6: 2024 46

INTRODUCTION

The synaptic connections between corticospinal axons 
and motor neurons in the spinal cord play a crucial role 
in transmitting signals essential for coordination, move-
ment, and sensory functions. Spinal Cord Injuries (SCIs) 
can damage the descending corticospinal axons, leading 
to the disruption or attenuation of nerve signals between 
the central nervous system (CNS) and the peripheral 
nervous system (PNS).1 

In the event of a stroke, an obstruction (ischemic stroke) 
or breakage of a blood vessel (hemorrhagic stroke) may 
result in brain damage to regions governing movement 
or sensation by inhibiting the production or transmission 
of neural signals. 

The weakening or interruption of neural connections in 
both cases is a condition known as denervation. Affected 
limbs or organs may experience a range of symptoms, 
from weakness and numbness to loss of sensory function 
and complete paralysis.2

Denervation and its symptoms can be mitigated through 
rehabilitation protocols designed for cortical reorganiza-
tion. These protocols are grounded in the principles of 
Hebbian learning and leverage the brain’s inherent capacity 
for adaptation, a phenomenon known as neuroplasticity.2 

Neuroplasticity is a term used to describe the CNS’s 
neurodevelopmental capability to experience alterations 
in structure and function, following exposure to both ex-
ternal and internal stimuli. This capacity is not exclusive to 
a specific time frame of human life. Hence, neuroplasticity 
is crucial, in that it facilitates healing in response to CNS 
injury and trauma for the entire duration of human life.3

Hebbian learning is based on the hypothesis that gains 
in synaptic efficacy are realized following the exertion of 
repeat stimulation of a postsynaptic cell by a presynaptic 
cell. It is a model of associative learning which proposes 
that synchronous neural activity produces increased 
synaptic strength in the cells involved.4

Electrical Muscle Stimulation (EMS) encompasses a 
wide array of therapeutic interventions. In general terms, 
electrical pulses are applied to the neuromuscular system 
offering additional stimuli for Nervous System activation 

and reorganization. The electric pulses elicit action poten-
tials that bring about muscle contractions capable of being 
synchronized with the functional movements and tasks 
performed during a rehabilitation session, a methodol-
ogy known as Functional Electrical Stimulation (FES).1,2

FES constitutes a tool capable of stimulating the neu-
romuscular system, thus occasioning neuromuscular and 
central nervous system plasticity.2 Due to the precise tim-
ing, it further leverages the principles of Hebbian learning 
enhancing synaptic connections and the formation of 
neural pathways. Employed in response to both strokes 
and spinal cord injuries, FES has been found to produce 
better outcomes with regard to patient mobility, spastic-
ity, walking speed, and spinal cord function recovery.1,3,4 

The significant parameters affecting the quality of 
muscular contractions and cortical reorganization are 
pulse intensity and width, frequency, as well as the time 
delay between pulses. Notably, these parameters are 
session- and subject-specific because they are affected 
by factors such as the type of waveform, the placement 
of electrodes, electrolyte concentration in the targeted 
muscle, the cleanliness of the skin area where the elec-
trode is placed, the adaptation of FES parameters across 
different rehabilitation sessions, and the synchronization 
between voluntary command and the muscular contraction 
which is actually induced.1,2 Small variations in param-
eter values can substantially alter the quality of induced 
muscular contractions as well as the longevity of cortical 
reorganization.2 It is important to note that although higher 
pulse intensities and frequencies elicit stronger contrac-
tions they may also introduce pain, discomfort, and skin 
irritation.2 Consequently, there’s a pressing need for an 
automated calibration process at each sessions’ outset, 
determining optimal parameter values while considering 
patient comfort. 

MATERIALS AND METHODS 

Hardware

The system consists of a commercially available PC, 
two microcontrollers, a gyroscoping accelerometer, a 
programmable waveform generator, an operational ampli-
fier, electrodes, and an oscilloscope for data acquisition. 

http://www.globalce.org
http://globalce.org
http://globalce.org


47 J Global Clinical Engineering Vol.6 Special Issue 6: 2024

Software

The 8-bit microcontrollers were programmed by use 
of the C programming language. 

The first microcontroller managed the MPU6050 
gyroscoping accelerometer to monitor the magnitude 
of the acceleration produced by the stimulated muscle 
in real time, providing constant feedback to the system. 

The second of the two controlled the AD9833 program-
mable signal generator enabling real-time adjustments for 
every stimulation parameter such as intensity, frequency, 
pulse width, shape of the waveform, and more. 

An AI expert system tuned on accelerometry data is 
being used for optimization purposes. Frequency is be-
ing swept across a preset range while monitoring muscle 
responses. 

Data transfer from microcontrollers to PC was facili-
tated via a USB connection. All results were visualized 
by use of appropriate Matlab scripts.

Experimental Set up

The interconnectivity of components is illustrated in 
the block diagram of Figure 1. 

FIGURE 1. Block diagram of the experimental setup.
      

The AD9833 waveform generator produces the electrical 
stimulation pulses, which are adjusted and controlled by 
the 8-bit microcontroller. The signal is amplified and then 
transmitted to the left-hand bicep muscle via electrodes. 
The electrodes are always identical and consistently 
positioned at the muscle belly after the implementation 
of a standard cleaning protocol. This procedure ensures 
repeatable and uniform measurements. The amplification 
process utilizes a conventional non-inverting operational 
amplifier circuit. 

The microcontrollers feature the Atmega328P single 
chip by Atmel, run on an 8-bit AVR processor core, and 
incorporate a 16 MHz quartz crystal oscillator. The chip 
is designed with 6 analog inputs, and 14 digital input/
output pins and offers 32 KB flash memory, 2 KB SRAM, 
1 KB EEPROM and a wired USB interface for program-
ming purposes. 

The MPU6050 gyroscoping accelerometer integrates 
a 3-axis gyroscope and a 3-axis accelerometer to endow 
low noise and precise 6-axis motion tracking. It operates 
within a supply voltage range of 2.375–3.46 V, has an 
adjustable range of ± 16 g, comes equipped with a Digital 
Motion Processor, and supports an I2C interface. 

AD9833 programmable waveform generator operates 
within a supply voltage range of 2.3–5.5 V and consumes 
20 mW. The device is capable of producing sinusoidal 
waveforms with peak-to-peak amplitude of 0.6 V as well 
as triangular and square pulses with peak-to-peak ampli-
tude equal to the supply voltage. The frequencies of the 
generated waveforms span from 0–12.5 MHz with 0.1 Hz 
accuracy. The chip supports communication via several 
protocols, including SPI (utilized in this instance), QSPI, 
and MICROWIRE. 

The OPA462IDDA SMT high voltage operational am-
plifier operates within a supply voltage range of ± 6 to 
± 90 V and can provide 30 mA current. It incorporates 
protection mechanisms against overheating and current 
overloads, is unity stable with a gain-bandwidth product 
of 6.5 MHz, a slew rate of 32 V/μs, and has a high output 
load drive of ± 45 mA. 

A set of commercially available rectangular-shaped 
(dimensions: 4.5 × 3.5) pre-gelated, self-adhesive trans-
cutaneous electrodes. 

The Picoscope 2000 series was used to monitor the 
waveforms generated by the AD9833. The device offers 
several triggering options, and boasts 200 MHz bandwidth, 
12-bit resolution, and 128 MS memory capacity. While the 
Picoscope has an integrated function generator, it wasn't 
employed in these particular experiments. 

http://www.globalce.org
http://globalce.org
http://globalce.org


J Global Clinical Engineering Vol.6 Special Issue 6: 2024 48

The other microcontroller oversees the MPU6050 
tracking the magnitude of acceleration of the stimulated 
muscle contraction during its concentric phase. The chip 
is affixed to a glove consistently worn on the left palm. 
The maximum magnitude of acceleration generated dur-
ing the concentric phase of each movement is captured 
and subsequently visualized in a graph to be utilized for 
optimization. 

The waveform employed is a symmetric biphasic 
square pulse maintaining a stable peak-to-peak voltage 
of 35 V. The muscle’s physical movement is assessed 
across varying frequencies, starting from a threshold of 
10 Hz and incrementing in steps of 5 Hz up to a ceiling of 
140 Hz. Then, the maximum magnitude of acceleration 
for each contraction is plotted against its corresponding 
frequency. Higher frequencies (and peak-to-peak volt-
ages) stimulate more motor units, resulting in stronger 
muscle contractions. However, there’s a threshold beyond 
which no additional motor units are engaged. It’s crucial 
to recognize that pushing these parameters to their limits 
isn’t the best approach as it leads to patient’s pain and 
discomfort, which can hinder the rehabilitation process. 
Examining the magnitude of acceleration as a function of 
frequency and determining the peaks of that function can 
provide optimal values of frequency that elicit stronger 
movements than higher frequencies. These frequencies 
can later be used for a better rehabilitation session by 
taking patients’ pain and discomfort levels into account. 

RESULTS 

The results showcased in Figures 2, 3, and 4 are derived 
from the implementation of the previously described 
optimization algorithm on the same subject over differ-
ent EMS sessions. 

Examination of the peaks of the maximum acceleration 
to frequency graph in Figure 3 reveals possible optimal 
frequencies at 70 Hz and at 120 Hz. In case of pain at 
higher frequencies, 70 Hz emerges as a more fitting re-
placement of the function’s global maximum at 120 Hz. It 
can be characterized as an optimal value for the specific 
rehabilitation session as it elicits comparatively stronger 
contractions than higher frequencies. 

FIGURE 2. Normalized Maximum Acceleration against frequency 
at a voltage differential of 35 V (session A).

Upon analyzing the peaks in the maximum acceleration 
to frequency graph in Figure 4, potential optimal frequen-
cies at 60 Hz, 80 Hz, 125 Hz, and 135 Hz are identified. 
Should higher frequencies induce pain, 60 Hz becomes 
the preferred choice. Conversely, 125 Hz can be utilized 
as optimal frequency adhering to the same logic. 

FIGURE 3. Normalized Maximum Acceleration of the electri-
cally stimulated muscle against frequency at a voltage of 35 V 
(session B).

Following the same reasoning and after analyzing 
the graph in Figure 4, potential optimal frequencies are 
identified at 50 Hz, 90 Hz, 110 Hz, 120 Hz, and 135 Hz.

http://www.globalce.org
http://globalce.org
http://globalce.org


49 J Global Clinical Engineering Vol.6 Special Issue 6: 2024

intramuscular electrolyte concentration, adaptive re-
sponses to stimuli, alterations in strength and coordination, 
and fat percentage in the adjacent area, among others. 
This underscores the necessity for an automated, short 
calibration process which computes the optimal value 
for frequency, voltage, and other EMS parameters at the 
beginning of each EMS session. 

Another challenge stemming from the variability of 
responses to EMS across sessions is the ability to com-
pare results both between subjects and across sessions. 
One potential solution is to highlight deviations from the 
mean response at varying frequencies by normalizing 
the y-axis of the graphs in units of standard deviation. 

In certain instances, determining an optimal frequency 
using this method may be challenging, especially if the 
data does not exhibit distinct local maxima. We expect this 
issue to be addressed by considering more parameters 
for optimization and by integrating Machine Learning 
techniques into our approach. 

DISCUSSION 

It should be highlighted that all results are preliminary. 
Experimentation with more subjects is required in order 
to draw more definitive conclusions. 

All necessary steps were taken to ensure compliance 
with the General Data Protection Regulation and appli-
cable national law. Any future development of products 
for commercial or other use stemming from this research 
will be governed by and will have to adhere to Regulation 
(EU) 2017/745 on medical devices and the General safety 
and performance requirements it establishes. 

CONCLUSION AND FUTURE WORK

The results seem promising although preliminary. Both 
conventional and AI-facilitated optimization methods 
demonstrate the potential to mitigate discomfort and 
muscle fatigue experienced during FES sessions and 
multiple optimization methods should be explored and 
compared. By addressing these challenges, FES will become 
usable outside of clinical trials as a tool for daily tasks, 

FIGURE 4. Maximum Acceleration against frequency at a volt-
age of 35 V (session C).

Considering the data as depicted in the graphs above, it 
becomes evident that the optimal value for frequency, as 
well as other parameters, as mentioned in the literature, 
is not only subject but also session-specific. Sensitivity 
to EMS stimulation, each subject’s perception of pain at 
different frequencies, and the muscle’s response vary 
between sessions. This is apparent both when examining 
optimal frequencies as well as when comparing the mean 
value and standard deviation of maximum acceleration 
for each session. Results are summarized in Table 1. 

TABLE 1. 

Session Mean Max Acceleration Std Deviation

A 19.59m/s^2 2.15m/s^2 

B 17.91m/s^2 3.02m/s^2 

C 18.88m/s^2 2.48m/s^2 

We interpret higher average maximum acceleration in 
session A as higher sensitivity to EMS for that particular 
session. The higher value of Standard Deviation in session 
B shows a higher sensitivity in frequency fluctuations, 
again for that particular session. The variation in values 
may depend on various factors, including skin cleanliness, 

http://www.globalce.org
http://globalce.org
http://globalce.org


J Global Clinical Engineering Vol.6 Special Issue 6: 2024 50

improving quality of life. Additionally, the rehabilitation 
process can reap significant benefits, paving the way for 
a swifter and pain-free recovery. 

In our future work, we plan to increase our number 
of subjects. We aim to incorporate Machine Learning 
Techniques into our Expert System AI, add an electrode 
array to our setup, and study different electrode activation 
patterns, positions, and parameter settings. 

REFERENCES

1. Karamian, B.A., Siegel, N., Nourie, B., et al. The role of 
electrical stimulation for rehabilitation and regeneration 
after spinal cord injury. J Orthop Traumatol. 2022;23(1):2. 
https://doi.org/10.1186/s10195-021-00623-6. 

2. Milosevic, M., Marquez-Chin, C., Masani, K., et al. Why 
brain-controlled neuroprosthetics matter: mechanisms 
underlying electrical stimulation of muscles and nerves 
in rehabilitation. BioMed Eng OnLine. 2020;19:81. 
https://doi.org/10.1186/s12938-020-00824-w.

3. Christiansen, L. and Siebner, H.R. Tools to explore 
neuroplasticity in humans: Combining interventional 
neurophysiology with functional and structural mag-
netic resonance imaging and spectroscopy. Handb Clin 
Neurol. 2022;184:105–119. https://doi.org/10.1016/
B978-0-12-819410-2.00032-1. 

4. Jo, H.J., Kizziar, E., Sangari, S., et al. Multisite Hebbian 
Plasticity Restores Function in Humans with Spinal 
Cord Injury. Ann Neurol. 2023;93(6):1198–1213. 
https://doi.org/10.1002/ana.26622.

http://www.globalce.org
http://globalce.org
http://globalce.org
https://doi.org/10.1186/s10195-021-00623-6
https://doi.org/10.1186/s12938-020-00824-w
https://doi.org/10.1016/B978-0-12-819410-2.00032-1
https://doi.org/10.1016/B978-0-12-819410-2.00032-1
https://doi.org/10.1002/ana.26622

	Editor’s Corner
	Biomedical Technology and Clinical Engineering in Greece after the Pandemic: Highlighted Works from the Panhellenic Conference of Biomedical Technology
	Aris Dermitzakis1,2,*, Vasiliki Zilidou1,3, Eleftheria Vellidou1,4, Alkinoos Athanasiou1,3

	Digital Transformation Management in Health Services: Health Professionals Perceptions as an Implementation Factor
	Theodoros S. Tanis*, Chryssoula Chatzigeorgiou, Ioanna Simeli, and Evangelia Stalika

	Validating the ID-GAMING e-Training Toolkit for People with Intellectual Disabilities in Greece
	Niki Pandria*, Anastasia Barboudi, Vasileia Petronikolou, Panagiotis Antoniou and Panagiotis D. Bamidis

	Novel Functional Electrical Stimulation Parameter Optimization for Neurorehabilitation Using Both Conventional and AI Techniques 
	Arsenios Arsenidis1, Alexandros Moraitopoulos2, Alkinoos Athanasiou2, Alexandros Vildiridis3, Panagiotis Bamidis2, Petros Stefaneas4 and Alexandros Astaras5 

	Leveraging Web Scraping and API Integration for Efficient Medical Device Data Management 
	Agapi Konstantina Liontou1,*, Spilios Zisimopoulos2 and Aris Dermitzakis1 

	Human Muscle State Machine Using Electromyography Classification with Machine Learning 
	George Lyssas1,*, Konstantinos Mitsopoulos1, Dimitris Zantzas2, Anestis Kalfas2, Panagiotis D. Bamidis1 

	Kinematic and Dynamic Analysis of Lower Limb Movement: Towards the Design of a Wearable Rehabilitation Assistant Device 
	Filippos Margaritis1,*, Konstantinos Mitsopoulos1, Kostas Nizamis2, Alkinoos Athanasiou1 and Panagiotis D. Bamidis1 

	A Novel Dermatological Diagnosis Support Device Based on Electrical Impedance Spectroscopy 
	Alexandros Moraitopoulos1,*, Konstantinos Mitsopoulos1, Christina Kemanetzi2, Panagiotis Bamidis1 and Alexandros Astaras3

	Software Skills Identification: A Multi-Class Classification on Source Code Using Machine Learning
	Dimitris Bamidis, Ilias Kalouptsoglou, Apostolos Ampatzoglou, Alexandros Chatzigeorgiou*

	Improvement of Aortic Valve Stenosis Classification in Patients Through Computational Fluid Dynamics Model
	Ioannis Makropoulos, Dimitris Zantzas, Vasilis Gkoutzamanis, Anestis Kalfas*

	Kinematic and Dynamic Analysis of the Human Hand’s Articulation for Wearable Soft-Robotic Device Applications
	Paschalina-Danai Sarra, Vasiliki Fiska, Konstantinos Mitsopoulos, Diamanto Mylopoulou, and Panagiotis D. Bamidis*

	Deep Learning Classification of Epileptic Magnetoencephalogram 
	Andreas Stylianou1, Lefteris Koumakis2, Maria Hadjinicolaou3, Adam Adamopoulos1,* and Alkinoos Athanasiou4


