Unveiling the Power of EEG Signals: Parkinson's Disease

Identification via Yet Another Mobile Network (YAMNet)

Ali Abdulameer Aldujaili

1,2 a

, Manuel Rosa-Zurera

2 b

and Ahmed Meri

3 c

Department Affairs of Student Accommodation, University of Baghdad, Baghdad, Iraq

Department of Signal Theory and Communication, University of Alcalá, Alcalá de Henares, Madrid, Spain

Department of Medical Instrumentation Techniques Engineering, Al-Hussain University College, Karbala, Iraq

Keywords: EEG, Parkinson’s Disease, Early Detection, Deep Learning, YAMNet.

Abstract: Parkinson's disease is a neurodegenerative disorder with a progressively debilitating impact on patients'

movement in terms of cognitive and motor aspects. Early detection is crucial for effective disease management

and better patient outcomes. There are many techniques to detect this disease, but one of the most interesting

methods to achieve early detection of Parkinson’s disease is electroencephalography, which is a non-invasive

and cost-effective diagnostic tool to measure brain activity. Recent studies have shown that deep learning

networks can handle complex data to analyse it and extract features. One of these neural networks is called

Yet Another Mobile Network (YAMNet), which was originally proposed to analyse speech signals using

time-frequency information. In this research, a novel approach using YAMNet is presented for the detection

of Parkinson's disease patients using electroencephalogram brain signals, as the frequency information seems

very relevant for Parkinson's disease detection. The proposed approach was evaluated with an open access

dataset available on the Internet, composed of electroencephalogram recordings from Parkinson's disease

patients and healthy control people, obtaining an accuracy rate of 98.9%. The results suggest that YAMNet

could be an encouraging tool for the initial, non-invasive detection of Parkinson's disease. This may improve

patient treatments and stimulate future research in the field.

1 INTRODUCTION

Parkinson's disease (PD) is a gradually progressive

neurodegenerative condition caused by the loss of

dopamine-producing cells in the brain, which leads to

motor and cognitive impairment (Zaman et al., 2021).

A diagnosis of Parkinson's disease ordinarily involves

an extensive assessment of the patient's medical

history, family history, and physical examination.

Bradykinesia, tremor, and rigidity are common

clinical manifestations in patients, and these are the

most prominent presenting symptoms (Balestrino &

Schapira, 2020). Furthermore, Parkinson's disease

has spread worldwide over time, increasing 2.4-fold

between 1990 and 2016 (Müller-Nedebock et al.,

2023).

On the other hand, the aetiology of the disease has

remained unknown until these days. Parkinson's

https://orcid.org/ 0000-0003-1114-5857

https://orcid.org/ 0000-0002-3073-3278

https://orcid.org/ 0000-0001-8873-1438

disease may present itself in diverse forms and cases,

each associated with distinct prognoses and disease

progressions (Lang & Espay, 2018). Moreover, a

malignant subtype is observed, and it represents a

small percentage of about 9% to 16% of patients. It is

characterized by swift disease advancement and by

the existence of both motor and non-motor

symptoms. On the contrary, a high rate of around 49%

to 53% of patients suffer from mild motor-dominant

Parkinson's, which progresses slowly and can be

effectively treated, reducing symptoms with

dopaminergic medications (Sabahi et al., 2021).

Research focused on early detection and

classification of the different subtypes is necessary to

determine the best treatment that can be offered to

patients. Researchers have investigated various

alternative diagnostic methods, such as handwriting

analysis, electroencephalography (EEG) signals

978

Aldujaili, A., Manuel, R. and Meri, A.

Unveiling the Power of EEG Signals: Parkinson’s Disease Identiﬁcation via Yet Another Mobile Network (YAMNet).

DOI: 10.5220/0012589100003654

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

In Proceedings of the 13th International Conference on Pattern Recognition Applications and Methods (ICPRAM 2024), pages 978-984

ISBN: 978-989-758-684-2; ISSN: 2184-4313

analysis, magnetic resonance imaging (MRI), voice

analysis, and movement tests, in order to obtain an

early diagnosis and detect PD (Tăuţan et al., 2021).

It is worth noting, for example, the approach of

using human gait movement patterns to

neurodegenerative diseases classification, by using

the kinematic theory of rapid human movements

(Dentamaro et al., 2020).

The same research group has also addressed the

problem of Alzheimer’s and Parkinson’s diseases

detection and classification, by the analysis of

handwritten trials from a pattern recognition

perspective, including data acquisition, feature

extraction, data analysis, and classification

(Impedovo & Pirlo, 2018).

The most important kinetic features and the most

significant tasks for neurodegenerative disease

assessment through handwritten have been identified

in (Dentamaro, Impedovo, et al., 2021), working with

the novel HAND-UNIBA dataset.

Regarding the application of machine learning for

neurogenerative diseases detection or classification, it

is worth mentioning that this group have addressed

the application of different machine learning tools,

including shallow learning techniques, and deep

learning with transfer learning, for neurodegenerative

disease assessment through handwriting (Gattulli et

al., 2022) (Dentamaro, Giglio, et al., 2021),

demonstrating that this approach based on

handwriting analysis combined with artificial

intelligence techniques, is useful for early detection

of neurodegenerative diseases.

Although there are many signals that can be used

for early detection of Parkinson’s disease, EEG

signals have gained much attention in recent times

due to their convenience of acquisition, cost-

effectiveness and high level of accuracy. In general,

EEG signals present energy in the band between 0 and

100 Hz, which can be divided into five sub bands:

delta, theta, alpha, beta and gamma (Khosla et al.,

2020). Traditionally, EEG signals are processed using

spectral analysis techniques, so the implementation of

automatic diagnostic systems is based on the

extraction of relevant features in the frequency

domain. The extracted features are used to feed

classifiers that perform the task of classification

between healthy and sick people, or between different

degrees of the disease, depending on what the

objective is.

Convolutional neural network (CNN) models

have been successfully applied as tools for feature

extraction from EEG signals and the subsequent

diagnosis, providing advantages over other types of

systems, such as automatic feature extraction,

improved accuracy, robustness, and real-time

analysis (Maitin et al., 2022). This paper proposes the

use of deep learning models fed by information of

EEG signals in the frequency domain and

demonstrates its suitability to detect Parkinson’s

disease.

Both classical machine learning and deep learning

techniques offer the possibility of early identification

of Parkinson's disease by analysing huge amounts of

EEG data (Maitín et al., 2020). However, deep

learning algorithms can identify minor variations in

brain activity that may not be detectable with

conventional diagnostic techniques. This allows

earlier diagnosis, creating more effective treatment

options. In addition, deep learning models can solve

problems such as noise, distortion and variation in

EEG signals due to various factors, such as electrode

placement, motion or magnetic interference. In

addition, these models can learn complex features

independently, thus reducing the need for manual

interference, which decreases reliance on subject

expertise and self-interpretation (Khan et al., 2021).

Research into EEG techniques is advancing, with

the aim of improving the accuracy and capabilities of

Parkinson's disease detection and monitoring.

Finally, the emergence of innovative approaches in

EEG analysis through deep learning shows

substantial potential for the diagnosis and

management of Parkinson's disease. These advances

have the potential to improve early detection,

establish superior treatment options, and

consequently improve outcomes for patients and their

families.

In this paper, a deep neural network proposed for

audio processing is applied to the detection of PD

using EEG signals. The neural network is known as

Yet Another Mobile Network (YAMNet) (Plakal, M.

& Ellis, 2020), and it is a pre-trained deep neural

network that can predict audio events from 521

classes. Audio and EEG signals share the

characteristic that the information is mainly in the

frequency domain, and we wonder if this pre-trained

network for audio classification could be fitted to

solve the problem of PD diagnosis, using the

internally extracted features for audio classification.

This is the hypothesis underlying the research in this

paper.

The paper is organised as follows. Section 1

contains the introduction to the problem the paper

deals with. Section 2 reviews the main related works.

Section 3 describes the materials and methods used in

the research. The results are presented and discussed

in Section 4. Finally, Section 5 contains the

conclusions.

Unveiling the Power of EEG Signals: Parkinson’s Disease Identiﬁcation via Yet Another Mobile Network (YAMNet)

979

2 RELATED WORKS

Various studies have been conducted to detect

Parkinson's disease (PD) by analysing patients’ EEG

signals with deep learning techniques. For instance,

Zhang et al (Zhang et al., 2022) employed the

tuneable Q-factor wavelet transform (TQWT),

wavelet packet transform (WPT), and deep residual

shrinkage network (DRSN) to attain high

classification accuracy. TQWT achieved 99.67%

accuracy with features Permutation Entropy (PE) and

Order Index (OI), whereas WPT produced 99.83%

accuracy with energy features, and DRSN exhibited

99.87% accuracy.

In another study, Lee et al (Lee et al., 2021) were

able to differentiate Parkinson's disease (PD) from

healthy controls with high accuracy (99.2%),

precision (98.9%), and recall (99.4%) using

convolutional-recurrent neural networks (CRNN).

They achieved this by extracting spatial and temporal

features in multi-channel EEG signals. However, it

was acknowledged that the model had certain

limitations, including sensitivity to medication

effects, a small sample size, and issues related to

model interpretability.

On the other hand, Xu et al. (Xu et al., 2020) used

a pooling-based deep recurrent neural network

(PDRNN), resulting in a precision of 88.31%, a

sensitivity of 84.84%, and a specificity of 91.81%.

Despite these results, there were cases of

misclassification, which involved 11.28% of healthy

patients being classified as Parkinson's sufferers and

11.49% of Parkinson's cases being incorrectly

identified as healthy. The research was also restricted

by a small number of participants and potentially high

computing costs when compared to conventional

machine-learning methods.

The chaos theory has been applied by Shah et al.

(Shah et al., 2020) to analyse variations in the EEG

patterns of Parkinson's disease (PD) patients with a

view to discovering a biomarker for PD classification.

For classification tasks, they used the CNN

Dynamical System Generated Hybrid Network

(DGHNet) and Long Short-Term Memory (LSTM)

units with the EEGLAB toolbox. Their study

achieved a remarkable 99.2% accuracy in the

classification of PD cases, even with limited

computational resources. However, it was observed

that additional research is required to address gaps in

inter-patient classification due to the complex and

patient-specific nature of EEG data.

Additionally, Oh et al. (Oh et al., 2020)

constructed a thirteen-layer CNN model for detecting

Parkinson's disease, attaining an accuracy rate of

88.25%, a sensitivity rate of 84.71%, and a specificity

rate of 91.77%. This study was restricted by the

sample size and the computational cost of the CNN

configuration.

Shi et al. (Shi et al., 2019), in another study,

utilized hybrid models composed of two conventional

deep learning models (CNN and RNN) to categorize

PD and normal EEG signals. The hybrid models

performed better than the conventional models;

however, the authors suspected that some data

included in the database might be mislabelled. In

addition, the amount of data from the PD and HC

groups was small, so the five-fold technique was used

to estimate the mean accuracy, obtaining the

following results: 3D- CNN-RNN 82.89%, 2D-CNN-

RNN 81.13%, CNN 80.89%, and RNN 76.00%.

Lee et al. (Lee et al., 2019), presented a

framework that combines a convolutional neural

network (CNN) and a recurrent neural network

(RNN) with LSTM cells. The proposed model

achieved remarkable outcomes, as it achieved an

accuracy of 96.9%, precision of 100%, and recall of

93.4%. Consequently, the framework readily

distinguishes PD from healthy controls. The

researchers suggest refining the model with a larger

dataset could make the CRNN framework a valuable

diagnostic tool for monitoring diseases.

In a different study by Loh et al. (Loh et al., 2021),

the authors investigated a deep-learning model for PD

analysis. By using the Gabor transform to convert

EEG recordings into spectrograms, they attained a

99.46% accuracy when training their 2D-CNN

model. Additionally, the authors stressed the

significance of broadening the model's capacity by

integrating information on other brain irregularities,

such as sleep disorders, depression, and autism, for

multiple brain disorder identification as opposed to

exclusively targeting one ailment.

An analysis of the literature revealed advantages

and disadvantages in using deep learning techniques

and algorithms with EEG brain signals to diagnose

Parkinson's disease in patients.

Additionally, the studies discussed had some

limitations, such as small sample sizes, expensive

computing requirements, and limited interpretability

of the proposed models, despite simulation studies

being conducted for their assessment. Additionally,

certain models require further enhancements to

ensure compatibility with cloud systems and detect

multiple ailments. Thus, it is imperative to conduct

more research to attain accurate identification of

Parkinson's disease from EEG signals using deep-

learning methodologies.

NeroPRAI 2024 - Workshop on Medical Condition Assessment Using Pattern Recognition: Progress in Neurodegenerative Disease and

Beyond

980

The main conclusion drawn from the literature

review is that EEG signals are useful for

distinguishing healthy people from patients with

Parkinson's disease, but that there is at the same time

a major problem related to the amount of data

available for deep network training.

To circumvent the problem of limited available

data, several techniques have been explored, among

which data augmentation and transfer learning are

worth mentioning. While the use of techniques to

synthetically augment the data set is problematic in

healthcare applications, transfer learning techniques

look promising.

Sufficiently large databases of EEG signals, even

if related to other problems, are not available to train

deep networks and then use the trained network to

adjust it for the problem at hand, which is PD

detection. For this reason, in this paper we explore a

new line of research, which is the use of a pre-trained

network for another very different problem, which is

tuned a posteriori to solve the PD detection problem.

We have chosen the YAMNet network, pre-trained

for audio event classification, to solve our problem,

since both audio classification and EEG processing

for PD detection are performed in the frequency

domain.

3 MATERIALS AND METHODS

3.1 Material

This study on Parkinson's disease used a dataset that

was sourced from the Open-Neuro website, which is

accessible to the public (Rockhill et al., 2020). Two

sets of data make up the dataset: 15 PD patients' EEG

recordings make up the first group, known as the PD

dataset, while sixteen healthy controls who were

taking dopaminergic medications both ON and OFF

make up the second group. On their first usage,

technical terms will be defined. The data was split

into training (80% of the total) and validation (the

remaining 20% chosen randomly), following the

previously mentioned procedures.

So as to process the signals, a deep-neural

network known as Yet Another Mobile Network

(YamNet) has been used. This kind of deep

convolutional neural network (CNN), has been

proposed for audio classification. YamNet extracts

relevant acoustic features from audio waveforms to

classify sound events, and can predict 521 classes of

audio events, after being trained on the AudioSet-

Youtube corpus using the depth wise-separable

convolution architecture Movilenet_v1. The

architecture of YamNet was specially selected for the

application of audio classification. This deep neural

network is composed of 86 layers (Mohammed et al.,

2023):

• 27 layers for convolutional operations, which

play a key role in extracting meaningful features

from the input audio waveforms.

• 27 layers for batch normalization, to ensure that

the data is properly normalized, avoiding training

challenges and improving learning speed.

• A ReLU activation function is incorporated after

normalization, thus controlling the

computational complexity of the network.

• The structure is completed with one average

pooling layer, one fully connected layer, one

softmax layer, and a final classification layer.

These layers consolidate the information from

previous layers.

YamNet exploits the transfer learning paradigm.

After being trained for audio classification with the

aforementioned dataset, it can be adapted with

different data to other problems.

The model has learned patterns and characteristics

associated with different sound events through

training on a diverse set of labelled audio data. The

audio signals are transformed to obtain Mel-

spectrograms, that are applied as images to the

network. Because of that, the input EEG signals are

stored in *.wav files, considering the sampling rate is

16kHz, which is common in audio signals.

Moreover, the Yamnet network was built using

Matlab 2022b software on a PC equipped with two

Intel Xeon 3.10 GHz (E5-2687W) processors, 128

GB of RAM, and a 6 GB graphics card. The EEG data

file from the dataset with the (.bdf) extension was

read with an EEG tool in Matlab. Upon reading the

file, the tool automatically converted the file format

to (.mat) to be used in Matlab. The data underwent

multiple processing stages.

3.2 Methods

The Yamnet process segments the wave audio into

small overlapped windows, and transforms each one

into a 96x64x1 Mel-spectrogram image to extract

features. The images were then categorized, labelled

and stored into two folders, one for the health control

group and another one for Parkinson's disease

Unveiling the Power of EEG Signals: Parkinson’s Disease Identiﬁcation via Yet Another Mobile Network (YAMNet)

981

patients without medication. Each folder contained a

total of 840,000 images.

In order to extract more features, the images were

converted from double-valued images to the image

domain of 256 colours and then converted into a

grayscale matrix. The subsequent stage involved

commencing the training process on the data.

Table 1 presents the initial values provided for the

primary training hyperparameters. No layers were

altered during the training process, just the last layer

in full connection to be compatible with the

classification of output.

The deep neural networks were trained with the

ADAM optimizer, which significantly reduces

training time while achieving remarkable results. The

neural network was trained on the given dataset for

fifty epochs. In order to avoid suboptimal results or

an excessively extended training process, it is

paramount that the learning rate be carefully

calibrated and not set too high or too low. In this

work, a learning rate of 1e-5 was deemed optimal.

Ultimately, the YAMNet neural network

classified the data into two categories: health control

and Parkinson's disease without medication, as shown

in Figure 1, through its model architecture and

general process steps.

Figure 1: Model architecture.

Table 1: Hyperparameters used during the training phase.

Hyperparameter

value

Optimizer

Adam

Initial Learning Rate

0.00001

Mini batch size

Max epochs

F1 Score

Validation frequency

4 RESULTS AND DISCUSSION

The YAMNET model has been applied for the first

time to the detection of Parkinson's disease through

EEG signals, with the aforementioned database

containing 31 files. All the signals in the files were

segmented, and Mel-spectrograms were obtained. For

training and validation, the dataset was split into a

subgroup with 80% of images for training, and the

remaining 20% for validation. With this strategy, the

model achieved a remarkable accuracy of 98.97%.

The model's performance was assessed, obtaining the

confusion matrix that is shown in Table 2, and the

results that appeared are shown in Table 2.

Table 2: Results of the confusion matrix.

Predicted Class

True Class

Healthy

control

249198

2802

2362

249638

Healthy control

Table 3: Results.

Evaluation Metrics

Value

Accuracy

98.97%

Recall

99.06%

Precision

98.89%

Specificity

98.89%

F1 Score

98.93%

The study yielded encouraging results, as the

confusion matrix showed the recognized samples

correctly highlighted with dark boxes. The training

process took 29,181 minutes and 28 seconds,

requiring 50 epochs and 5,880,000 rounds of

iterations.

5 CONCLUSIONS

In this work, we suggest an EEG-based approach to

detect Parkinson's disease (PD) at an early stage. We

process the EEG signals using YAMNet as input

sound waves. The results show that the EEG signals

of the two groups under study (PD and HC) can be

classified with a high diagnostic accuracy, reaching

up to 98.97% accuracy. These results are comparable

to the best published in the literature but have been

obtained using a model pre-trained to solve a very

different problem, using transfer learning to tune the

model for the problem at hand, that is PD detection.











 

NeroPRAI 2024 - Workshop on Medical Condition Assessment Using Pattern Recognition: Progress in Neurodegenerative Disease and

Beyond

982

This confirm that transfer learning is useful, even

with networks trained for a very different problem,

opening a new strategy to overcome the problem of

the lack of samples for training in heath related

applications.

Detecting PD in its early stages is one of the

challenges facing the world, as it is essential for

successful treatment and improving patients' quality

of life, and we must race against time to find ways to

treat this dangerous disease. The YAMNet-based

method offers an affordable and non-invasive

solution for the detection of Parkinson's disease, thus

reducing examination time and workload in hospitals

and healthcare centres. In addition, it can be used in

real time, allowing PD patients to be continuously

monitored using wearable technology, and to

diagnose the patient earlier to offer tailored treatment

options. It should be noted that the study had

limitations, including the small sample size, which

made training and testing more complicated and

required the use of the k-fold technique, and the fact

that the database belonged to a restricted age group.

Moreover, in order to enhance diagnostic

precision and formulate a model for identifying and

categorizing other diseases that are also related to

analysing brain signals, it is crucial to verify the

efficacy of our technique with a wider range of

population samples and a larger dataset and variety.

Additionally, the implementation of this approach on

wearable devices to enable continuous monitoring of

PD patients poses several challenges that call for

further research.

ACKNOWLEDGEMENTS

This paper is part of the project with reference

PID2021-129043OB-I00, funded by

MCIN/AEI/10.13039/501100011033/FEDER, EU.

REFERENCES

Balestrino, R., & Schapira, A. H. V. (2020). Parkinson

disease. European Journal of Neurology, 27(1), 27–42.

https://doi.org/10.1111/ene.14108

Dentamaro, V., Giglio, P., Impedovo, D., & Pirlo, G.

(2021). Benchmarking of shallow learning and deep

learning techniques with transfer learning for

neurodegenerative disease assessment through

handwriting. International Conference on Document

Analysis and Recognition, 7–20.

Dentamaro, V., Impedovo, D., & Pirlo, G. (2020). Gait

Analysis for Early Neurodegenerative Diseases

Classification through the Kinematic Theory of Rapid

Human Movements. IEEE Access, 8, 193966–193980.

https://doi.org/10.1109/ACCESS.2020.3032202

Dentamaro, V., Impedovo, D., & Pirlo, G. (2021). An

analysis of tasks and features for neuro-degenerative

disease assessment by handwriting. International

Conference on Pattern Recognition, 536–545.

Gattulli, V., Impedovo, D., Pirlo, G., & Semeraro, G.

(2022). Early Dementia Identification: On the Use of

Random Handwriting Strokes. International

Graphonomics Conference, 285–300.

Impedovo, D., & Pirlo, G. (2018). Dynamic Handwriting

Analysis for the Assessment of Neurodegenerative

Diseases: A Pattern Recognition Perspective. IEEE

Reviews in Biomedical Engineering, 12, 209–220.

https://doi.org/10.1109/RBME.2018.2840679

Khan, P., Kader, M. F., Islam, S. M. R., Rahman, A. B.,

Kamal, M. S., Toha, M. U., & Kwak, K. S. (2021).

Machine Learning and Deep Learning Approaches for

Brain Disease Diagnosis: Principles and Recent

Advances. IEEE Access, 9, 37622–37655.

https://doi.org/10.1109/ACCESS.2021.3062484

Khosla, A., Khandnor, P., & Chand, T. (2020). A

comparative analysis of signal processing and

classification methods for different applications based

on EEG signals. Biocybernetics and Biomedical

Engineering, 40(2), 649–690.

https://doi.org/10.1016/j.bbe.2020.02.002

Lang, A. E., & Espay, A. J. (2018). Disease Modification

in Parkinson’s Disease: Current Approaches,

Challenges, and Future Considerations. Movement

Disorders, 33(5), 660–677. https://doi.org/10.1002/

mds.27360

Lee, S., Hussein, R., & McKeown, M. J. (2019). A deep

convolutional-recurrent neural network architecture for

Parkinson’s disease EEG classification. GlobalSIP

2019 - 7th IEEE Global Conference on Signal and

Information Processing, Proceedings, 14–17.

https://doi.org/10.1109/GlobalSIP45357.2019.896930

Lee, S., Hussein, R., Ward, R., Jane Wang, Z., &

McKeown, M. J. (2021). A convolutional-recurrent

neural network approach to resting-state EEG

classification in Parkinson’s disease. Journal of

Neuroscience Methods, 361, 109282.

https://doi.org/10.1016/j.jneumeth.2021.109282

Loh, H. W., Ooi, C. P., Palmer, E., Barua, P. D., Dogan, S.,

Tuncer, T., Baygin, M., & Rajendra Acharya, U.

(2021). Gaborpdnet: Gabor transformation and deep

neural network for parkinson’s disease detection using

eeg signals. Electronics (Switzerland), 10(14).

https://doi.org/10.3390/electronics10141740

Maitín, A. M., García-Tejedor, A. J., & Muñoz, J. P. R.

(2020). Machine learning approaches for detecting

parkinson’s disease from eeg analysis: A systematic

review. Applied Sciences (Switzerland), 10(23), 1–21.

https://doi.org/10.3390/app10238662

Maitin, A. M., Pablo, J., & Muñoz, R. (2022). applied

sciences Survey of Machine Learning Techniques in the

Analysis of EEG Signals for Parkinson ’ s Disease : A

Systematic Review. Applied Sciences (Switzerland),

Unveiling the Power of EEG Signals: Parkinson’s Disease Identiﬁcation via Yet Another Mobile Network (YAMNet)

983

12(14), 6967. https://doi.org/https://doi.org/10.3390/

app12146967

Mohammed, K. K., El-Latif, E. I. A., El-Sayad, N. E.,

Darwish, A., & Hassanien, A. E. (2023). Radio

frequency fingerprint-based drone identification and

classification using Mel spectrograms and pre-trained

YAMNet neural. Internet of Things (Netherlands),

23(July), 100879.

https://doi.org/10.1016/j.iot.2023.100879

Müller-Nedebock, A. C., Dekker, M. C. J., Farrer, M. J.,

Hattori, N., Lim, S. Y., Mellick, G. D., Rektorová, I.,

Salama, M., Schuh, A. F. S., Stoessl, A. J., Sue, C. M.,

Tan, A. H., Vidal, R. L., Klein, C., & Bardien, S.

(2023). Different pieces of the same puzzle: a

multifaceted perspective on the complex biological

basis of Parkinson’s disease. Npj Parkinson’s Disease,

9(1). https://doi.org/10.1038/s41531-023-00535-8

Oh, S. L., Hagiwara, Y., Raghavendra, U., Yuvaraj, R.,

Arunkumar, N., Murugappan, M., & Acharya, U. R.

(2020). A deep learning approach for Parkinson’s

disease diagnosis from EEG signals. Neural Computing

and Applications, 32(15), 10927–10933.

https://doi.org/10.1007/s00521-018-3689-5

Plakal, M. & Ellis, D. (2020). YAMNet.

https://github.com/tensorflow/models/tree/master/rese

arch/audioset/yamnet, Last accessed: 2023-12-19.

Rockhill, A. P., Jackson, N., George, J., Aron, A., & Swann,

N. C. (2020). uc san diego resting state eeg data from

patients with Parkinson’s disease. Available:

Sabahi, M., Joshaghanian, A., Dolatshahi, M., Jabbari, P.,

Rahmani, F., & Rezaei, N. (2021). Modification of

Glial Cell Activation through Dendritic Cell

Vaccination: Promises for Treatment of

Neurodegenerative Diseases. Journal of Molecular

Neuroscience, 71(7), 1410–1424.

https://doi.org/10.1007/s12031-021-01818-6

Shah, S. A. A., Zhang, L., & Bais, A. (2020). Dynamical

system based compact deep hybrid network for

classification of Parkinson disease related EEG signals.

Neural Networks, 130, 75–84.

https://doi.org/10.1016/j.neunet.2020.06.018

Shi, X., Wang, T., Wang, L., Liu, H., & Yan, N. (2019).

Hybrid convolutional recurrent neural networks

outperform CNN and RNN in Task-state EEG detection

for parkinson’s disease. 2019 Asia-Pacific Signal and

Information Processing Association Annual Summit

and Conference, APSIPA ASC 2019, November, 939–

944.

https://doi.org/10.1109/APSIPAASC47483.2019.9023

190

Tăuţan, A. M., Ionescu, B., & Santarnecchi, E. (2021).

Artificial intelligence in neurodegenerative diseases: A

review of available tools with a focus on machine

learning techniques. Artificial Intelligence in Medicine,

117(February).

https://doi.org/10.1016/j.artmed.2021.102081

Xu, S., Wang, Z., Sun, J., Zhang, Z., Wu, Z., Yang, T., Xue,

G., & Cheng, C. (2020). Using a deep recurrent neural

network with EEG signal to detect Parkinson’s disease.

Annals of Translational Medicine, 8(14), 874–874.

https://doi.org/10.21037/atm-20-5100

Zaman, V., Shields, D. C., Shams, R., Drasites, K. P.,

Matzelle, D., Haque, A., & Banik, N. L. (2021).

Cellular and molecular pathophysiology in the

progression of Parkinson’s disease. Metabolic Brain

Disease, 36(5), 815–827. https://doi.org/10.1007/

s11011-021-00689-5

Zhang, R., Jia, J., & Zhang, R. (2022). Biomedical Signal

Processing and Control EEG analysis of Parkinson ’ s

disease using time – frequency analysis and deep

learning. Biomedical Signal Processing and Control,

78(June), 103883.

https://doi.org/10.1016/j.bspc.2022.103883

NeroPRAI 2024 - Workshop on Medical Condition Assessment Using Pattern Recognition: Progress in Neurodegenerative Disease and

Beyond

984