% IMPORTANT: The following is UTF-8 encoded. This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.
@INPROCEEDINGS{Kuhles:1044377,
author = {Kuhles, Gianna and Thies, Tabea and Barbe, Michael and
Eickhoff, Simon and Camilleri, Julia and Weis, Susanne},
title = {{P}rediction of depression in {P}arkinson’s disease by
prosodic speech features},
reportid = {FZJ-2025-03152},
year = {2025},
note = {This study was supported by▪ the Deutsche
Forschungsgemeinschaft (DFG, GE 2835/2–1, EI 816/16-1 and
EI 816/21-1),▪ the National Institute of Mental Health
(R01-MH074457),▪ the Helmholtz Portfolio Theme
"Supercomputing and Modeling for the HumanBrain",▪ the
Virtual Brain Cloud (EU H2020, no. 826421) $\&▪$ the
National Institute on Aging (R01AG067103)},
abstract = {Introduction: Parkinson’s disease (PD) is a progressive
neurodegenerative disease whichmainly affects functions of
the motor system. Depressive disorders (DD) are one of the
mostcommon comorbidities in PD (Perez-Toro et al., 2022) and
often lead to a decrease in qualityof life (Balestrino $\&$
Martinez-Martin, 2017). Previous research revealed a strong
relationshipbetween prosodic impairment and DD (Vélez
Feijó et al., 2008) which is why recent studieshave
investigated whether speech parameters might be suitable as
biomarkers for the predictionof depression in PD. However,
the variability in feature extraction tools and
methodologies hasleft the most predictive prosodic
parameters inadequately defined (Perez-Toro et al., 2022).
Inour study, we aim to address this gap by identifying
reliable, standardised features for predictingdepression in
PD.Methods: Healthy subjects (n = 249) and people with PD (n
= 121) performed a spontaneousspeech task of 90 s duration.
88 acoustic features related to prosodic functions
wereautomatically extracted from these speech samples using
the toolbox OpenSMILE. Participantsalso completed the Beck
Depression Inventory (BDI-II) to assess depressive
symptoms.Prosodic features were used to classify between
healthy subjects and people with PD.Additionally, a machine
learning regression approach was employed to predict
individual BDI-II scores from prosodic speech features and
to determine the most significant predictors.Results:
Firstly, the classification analysis demonstrated a high
level of accuracy (0.997) indistinguishing between healthy
subjects and individuals with PD based on the prosodic
features.Secondly, the regression analysis revealed an R2
value of 0.024, indicating limited predictivepower for
BDI-II scores derived from prosodic features. However,
consistent with priorresearch, spectral parameters emerged
as the most significant predictors, especially the
MelFrequency Cepstral Coefficients, which describe the
perceived pitch of the frequency spectrum.Discussion:
Although the model demonstrated satisfactory performance in
classifying betweenhealthy individuals and those with
Parkinson's disease, the rather low accuracy in
predictingdepressive disorders (DD) highlights the
complexity of using prosodic features for this purpose.While
present results need to be validated in larger and diverse
samples, they pave the way tothe development of uniform
methodologies for the systematic extraction of
speech-relatedbiomarkers in PD, which in the long run might
have clinical utility for better understanding DDin people
with PD.Pérez-Toro PA et al. Speech Commun. (2022)
145:10-20.Balestrino R $\&$ Martinez-Martin P J. Neurol.
Sci. (2017) 374: 3-8.Vélez Feijó A et al. Neuro. Dis.
Treat. (2008) 4:669-674.},
month = {Jan},
date = {2025-01-26},
organization = {43. European Workshop on Cognitive
Neuropsychology, Bressanone (Italy), 26
Jan 2025 - 30 Jan 2025},
subtyp = {After Call},
cin = {INM-7},
cid = {I:(DE-Juel1)INM-7-20090406},
pnm = {5251 - Multilevel Brain Organization and Variability
(POF4-525)},
pid = {G:(DE-HGF)POF4-5251},
typ = {PUB:(DE-HGF)24},
doi = {10.34734/FZJ-2025-03152},
url = {https://juser.fz-juelich.de/record/1044377},
}