% 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{Amunts:887932,
author = {Amunts, Julia and Camilleri, Julia and Heim, Stefan and
Eickhoff, Simon and Weis, Susanne},
title = {{M}achine learning predicts executive functions from verbal
fluency data: {A} study in healthy participants},
school = {HHU Düsseldorf},
reportid = {FZJ-2020-04524},
year = {2020},
abstract = {Introduction: Executive functions (EFs) have been shown to
be crucial in performing the semantic verbal fluency (sVF)
task which is part of several neuropsychological test
batteries. While inhibitory control and working memory are
needed to monitor already produced words and prevent
repetitions, cognitive flexibility contributes to creative
thinking. By studying the role of clustering and switching
between words using manual approaches previous research
aimed to better understand participants’ search strategies
and the link between VF performance and EFs. The present
study makes use of a computerized analysis of both semantic
information and latency patterns to investigate whether
these variables can predict EF performance in healthy
individuals.Methods:Healthy monolingual German speaking
participants (n=234) completed 13 EF tests covering all
subdomains of EFs. Additionally, participants performed two
simple and one switching sVF task. The semantic information
was extracted using GermaNet [1], an ontological approach of
measuring semantic distances. Speech breaks and latency
patterns were determined with PRAAT [2]. A Relevance Vector
Machine (RVM) [3] approach with 100 replications and a
10-fold cross-validation was then employed to predict EF
scores from 30 sVF features adjusted for age and sex.
Prediction performance was assessed by the correlation
between true and predicted values. Features which had a
non-zero weight in at least $80\%$ of the models were
identified as the most influencing sVF variables.Results:The
prediction analysis revealed that sVF features significantly
predicted 7 EF tests tapping into all EF subdomains.
Specifically, latency patterns, especially those coming from
the first quarter of the sVF task, were identified as
predictive features. Moreover, the mean semantic relatedness
within each sVF task was identified as a meaningful feature
to predict EF scores. In particular, in the sVF switching
task the mean semantic relatedness was shown to reflect
inhibition performance. Results additionally show that
category errors could be linked to cognitive flexibility and
working memory performance.Discussion:The use of a machine
learning approach in the current study enabled an insight
into the complex and non-linear relationship between EFs and
the sVF task. In line with previous literature identifying
semantic distances as a sensitive measurement in
Alzheimer´s patients and latencies as a sensitive
measurement in ageing, our results reveal a specific link
between the switching sVF task and EF performance. They
highlight the potential of predicting EF scores from sVF
data by considering the combination of both semantic
information and latency patterns. Thus, combining
computerized speech parameter extraction and prediction
analyses carries the potential for an objective and
time-efficient alternative to manual evaluations in
assessing EFs from VF tasks in scientific and clinical
contexts.},
month = {Jan},
date = {2020-01-26},
organization = {European Workshop on Cognitive
Neuropsychology, Brixen (Italy), 26 Jan
2020 - 31 Jan 2020},
subtyp = {After Call},
cin = {INM-7 / INM-1},
cid = {I:(DE-Juel1)INM-7-20090406 / I:(DE-Juel1)INM-1-20090406},
pnm = {572 - (Dys-)function and Plasticity (POF3-572)},
pid = {G:(DE-HGF)POF3-572},
typ = {PUB:(DE-HGF)24},
url = {https://juser.fz-juelich.de/record/887932},
}