Assessing Parkinson’s Disease Speech Signal Generalization of

Clustering Results across Three Countries: Findings in the

Parkinson’s Voice Initiative Study

Athanasios Tsanas

and Siddharth Arora

Usher Institute, Edinburgh Medical School, University of Edinburgh, U.K.

Department of Mathematics, University of Oxford, U.K.

Keywords: Acoustic Analysis, Clustering, Parkinson’s Disease, Parkinson’s Voice Initiative (PVI).

Abstract: Progress in exploring speech and Parkinson’s Disease (PD) has been hindered due to the use of different

protocols across research labs/countries, single-site studies with relatively small numbers, and no external

validation. We had recently reported on the Parkinson’s Voice Initiative (PVI), a large study where we

collected 19,000+ sustained vowel phonations (control and PD groups) across seven countries, under

acoustically non-controlled conditions. In this study, we explored how well findings generalize in the three

English-speaking PVI cohorts (data collected in Boston, Oxford, and Toronto). We acoustically characterized

each sustained vowel /a/ phonation using 307 dysphonia measures which had previously been successfully

employed in speech-PD applications. We used the previously identified feature subset from the Boston cohort

and explored hierarchical clustering with Ward’s linkage combined with 2D-data projections using t-

distributed stochastic neighbor embedding to facilitate visual exploration of PD subgroups. Furthermore, we

computed feature weights using LOGO to assess feature selection consistency towards differentiating PD

from controls. Overall, findings are very consistent across the three cohorts, strongly suggesting the presence

of four main PD clusters, and consistent identification of key contributing features. Collectively, these

findings support the generalization of sustained vowels and robustness of the presented methodology across

the English-speaking PVI cohorts.

1 INTRODUCTION

Parkinson’s Disease (PD) is a crippling progressive

neurodegenerative disorder straining national health

systems due to increasing prevalence rates (Dorsey et

al., 2013). Indicatively, there were approximately 2.5

million People diagnosed with PD (PwP) in 1990, and

6.1 million PwP compared in 2016 (GBD, 2018).

Characteristic PD symptoms include tremor, rigidity,

bradykinesia, and postural stability, within the

broader remit of motor, cognitive, and

neuropsychiatric symptoms (Olanow, Stern, Sethi

2009). Similarly to some other disorders where a

disease name is used as an umbrella term, PD is well

reported as a largely heterogeneous disease with

considerable heterogeneity in PwP’s symptom

severity trajectories (Fereshtehnejad et al., 2015).

https://orcid.org/0000-0002-0994-8100

https://orcid.org/0000-0001-6499-6941

Exploring PwP phenotypes is clinically important

since homogeneous groups exhibit stronger clinical

symptom manifestation and potentially stronger

genetic coherence. In practice, PwP may be assigned

to specific subgroups based on clinical observations

and criteria such as age onset and dominating

symptoms. More recently, data-driven clustering

approaches have been explored to delineate PwP

subtypes using different data modalities. Indicatively,

research work has focused on clinico-pathological

characteristics (Selikhova et al., 2009), standardized

clinical instruments to assess motor, non-motor, and

cognitive domains (Lawton, 2018; Zhang et al.,

2019), or sensor-based gait pattern analysis (Nguyen

et al. 2019). The use of different types of data to

assess symptoms may provide new insights towards a

more holistic understanding of PD, however, makes

comparisons across studies particularly challenging

124

Tsanas, A. and Arora, S.

Assessing Parkinson’s Disease Speech Signal Generalization of Clustering Results across Three Countries: Findings in the Parkinson’s Voice Initiative Study.

DOI: 10.5220/0010383001240131

In Proceedings of the 14th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2021) - Volume 4: BIOSIGNALS, pages 124-131

ISBN: 978-989-758-490-9

and may explain discrepancies in the reported PD

subtypes and the dreaded replicability crisis in

science.

Ever since the detailed description of PD

symptoms by James Parkinson’s seminal work in

1817, speech has been known to be strongly affected.

In fact, 29% of PwP consider it one of their most

debilitating symptoms (Hartelius and Svensson,

1994). Recent studies have demonstrated the

enormous potential of capitalizing on speech signals

in neurodegenerative applications and PD in

particular. For example, research work has explored:

(1) differentiating PwP from age- and gender-

matched controls with almost 99% accuracy (Tsanas

et al., 2012), (2) accurately replicating the Unified

Parkinson’s Disease Rating Scale (UPDRS) (Tsanas

et al., 2011; Tsanas et al., in press), which is the

standard clinical tool to provide an overall PD

symptom assessment, and (3) automatically assessing

voice rehabilitation (Tsanas et al., 2014a). More

recently we have reported on the potential of speech

signals towards distinguishing people with Leucine-

Rich Repeat Kinase 2 (LRRK2) associated PD,

idiopathic PD, and control participants (Arora et al.,

2018). Similarly, speech articulation kinematic

models to characterize PD dysarthria have been

developed, which provide tentative insights into the

underlying physiology (Gomez et al., 2019).

Most studies in speech-PD report on single-site

findings, and are often limited in terms of the

statistical power due to the limited number of

available recordings, or the requirement of relying on

highly specific equipment and/or highly controlled

acoustic conditions. Motivated by the promising

findings we and others have reported in this field, we

set up a large multi-site trial and recently reported on

the Parkinson’s Voice Initiative (PVI) (Arora,

Baghai-Ravary, Tsanas, 2019). The PVI is a unique,

first of its kind, study where people were self-selected

and enrolled to participate, donating their voices

collected under acoustically uncontrolled conditions

over the phone. Overall, we have collected more than

19,000 sustained vowel /a/ samples from people

across seven countries. Although the data collected in

this study is clearly not of the same high quality as

data collected under carefully controlled acoustic

conditions, the large number of samples facilitates

new explorations in different directions.

The application of clustering algorithms using

speech signals has barely been explored. Rueda and

Krishnan (2018) used sustained vowel /a/ recordings

from 57 PwP and 57 matched controls to determine

groupings. However, the very small sample size

limits exploration and besides, mixing PwP with

controls is fundamentally not addressing the aim of

computing PD subtypes. Thus, to the best of our

knowledge we were the first to recently propose

clustering using sustained vowels to explore PwP

groupings (Tsanas and Arora, 2020). We had

previously used only the largest cohort (out of seven

cohorts) in the PVI to explore whether it is possible

to find some meaningful way to cluster PwP. The next

logical step is to validate how well those findings

generalize across other cohorts, which would

implicitly serve to assess the generalization of the

PVI project.

Therefore, the aim of this study is to explore how

well findings generalize across the three English-

speaking cohorts in PVI towards: (1) the computed

PwP clusters and (2) consistency of feature set

towards differentiating PwP from controls. The end

goal is to investigate whether the collected sustained

vowel /a/ phonations and proposed methodology has

internal consistency across different PVI datasets.

2 DATA

The PVI study invited people to self-enrol and

contribute their voices to facilitate clinical research in

PD. Data were collected across seven major

geographical locations (Argentina, Brazil, Canada,

Mexico, Spain, USA, and the UK) using servers by

Aculab for the needs of this project. People called a

dedicated phone number that was closest to their

geographical location and were requested to provide

some basic demographic information (age, gender),

self-report whether they had been clinically

diagnosed with PD, and record two sustained vowel

/a/ phonations. The instruction was to sustain vowel

/a/ for as long and as steadily as possible, following

standard widely used protocols which are easy to

implement (Titze, 2000). The speech recordings were

sampled at 8 kHz. In total, we collected more than

19,000 samples.

In this study we processed data from the three

English-speaking sites: Boston, Oxford, Toronto,

since we wanted to assess how well findings

generalize. Demographic information for the study

participants is summarized in Table 1; we do not have

detailed information regarding PD-symptom specific

aspects, for example whether participants self-

enrolled when they were “on” or “off” medication, or

clinically validated metrics such as UPDRS. For

further details on PVI we refer readers to our previous

work (Arora, Baghai-Ravary, Tsanas, 2019; Tsanas

and Arora, 2019).

Assessing Parkinson’s Disease Speech Signal Generalization of Clustering Results across Three Countries: Findings in the Parkinson’s

Voice Initiative Study

125

Table 1: Summary of demographics per cohort.

Boston Oxford Toronto

Participants

12171

(PD: 1138)

2103

(PD 285)

792

(PD: 107)

Phonations

12171

(PD:2097)

3908

(PD:536)

1461

(PD: 198)

63.7±10.8 63.5±10.0 65.0±9.8

Gender (males)

605 172 62

Distributions are summarized in the form mean ± standard

deviation. The basic demographic information is provided for the

PD participants since that is the main focus of the study.

3 METHODS

3.1 Data Pre-processing

We developed a speech recognition software which

automatically transcribed the participants’ responses

over the phone regarding age, gender, and self-

reported PD assessment. We aurally inspected

recordings where the automated speech recognition

algorithm had less than 90% confidence.

Furthermore, we developed an automated tool to

screen out unusable recordings, for example in the

presence of excessive background noise. For further

details please see (Arora, Baghai-Ravary, Tsanas,

2019).

3.2 Acoustic Characterization of

Sustained Vowel /a/ Phonations

We used the Voice Analysis Toolbox (freely

available from https://www.darth-group.com/

software) to acoustically characterize each sustained

vowel /a/ phonation. The toolbox computes 307

dysphonia measures, which have been developed

specifically to characterize sustained vowel /a/

phonations extensively validated across diverse PD

datasets (Tsanas et al., 2010a; Tsanas et al., 2010b;

Tsanas et al., 2011; Tsanas et al., 2012; Tsanas, 2012;

Tsanas et al., 2014a; Arora, Baghai-Ravary, Tsanas,

2019; Tsanas et al., in press), and other applications,

e.g. processing voice fillers (Tsanas and Gomez-

Vilda, 2013; San Segundo, Tsanas, Gomez-Vilda,

2017). We have described in detail previously the

background, rationale, and detailed algorithmic

expressions for the computation of the dysphonia

measures (Tsanas, 2012; Tsanas, 2013). A

prerequisite for the computation of many dysphonia

measures is the fundamental frequency (F0)

estimation. There are many algorithms in the research

literature for F0 estimation in different applications

(Tsanas et al., 2014b); here, we used the SWIPE

algorithm (Camacho and Harris, 2008), which we had

previously demonstrated is the most accurate F0

estimation algorithm in sustained vowel /a/

phonations (Tsanas et al., 2014b).

Applying the dysphonia measures to each

recording gives rise to features which are continuous

random variables. We linearly scaled each feature to

be in the range [0, 1] following standard practice for

distance-based machine learning algorithms so that

no feature dominates others (Bishop, 2006).

3.3 Feature Selection

A high dimensional dataset may obscure deciphering

of its core data structure and is typically challenging

for statistical learning algorithms. This well-known

problem is often referred to as the curse of

dimensionality, and may lead to detrimental

generalization of statistical learning algorithms

(Guyon et al. 2006; Hastie, Tibshirani, Friedman,

2009). Following Occam’s razor, we would prefer a

predictive model which is as simple as possible, i.e.

with a low dimensionality. This approach is known as

dimensionality reduction, and can be achieved either

by feature transformation (transforming the features

to populate a new, lower dimensional space), or by

feature selection (choosing a subset of features).

Feature selection is often more suitable in clinical

settings to retain the interpretability of the original

features (Guyon et al., 2006; Tsanas, Little,

McSharry, 2013).

There are two approaches in feature selection:

supervised (where the outcome information is used),

and unsupervised (where we do not have the

outcomes, or may not want to use that information).

Feature selection in unsupervised learning setups is

less studied and practically more challenging in terms

of defining a loss function (or criterion) to optimize

(Dy and Brodley, 2004). In this study we used both

unsupervised feature selection and supervised feature

selection to tackle the two different tasks.

For unsupervised feature selection, we used the i-

Detect to select informative features where the

identified feature subspace has the following

property: the difference between the total volume of

the space spanned by the selected feature subset and

the sum of the volumes of clusters in the embedded

manifolds is maximized (Yao et al., 2015). The i-

Detect algorithm has two free hyper-parameters: the

kernel width and the regularization parameter. The

algorithm is not very sensitive to the choice of the

kernel width (Yao et al. 2015), and hence we

experimentally explored the effect of optimizing the

regularization parameter. The output of i-Detect is a

BIOSIGNALS 2021 - 14th International Conference on Bio-inspired Systems and Signal Processing

126

sparse feature weight vector. The final ranking is

determined by the descending order of the weights.

For supervised feature selection, we used LOGO

(Sun et al., 2010), a feature weighting algorithm

which implicitly also provides an estimate of the

“importance” of each feature. Then, we determined a

minimal threshold and selected features in

descending order on the basis of decreasing feature

weights.

3.4 Clustering

Clustering is an unsupervised learning approach,

which attempts to group samples using the underlying

concept of sample distances. It can often provide

insight into the underlying structure of the data via the

(probabilistic) cluster membership of each sample

into the automatically determined clusters. Given

there are no labels (objective ground truth), clustering

is inherently more difficult to assess compared to

statistical learning models in supervised learning

setups.

Here, we used hierarchical clustering which is a

popular cluster analysis method that has been

successfully used across different applications

(Hastie, Tibshirani, Friedman, 2009). Hierarchical

clustering has a major advantage over some of the key

competing clustering approaches that it does not

require pre-specifying the number of clusters in the

data. Instead, it inherently constructs a dendrogram to

represent the data in a tree-based form, where the tree

is recursively split to form new clusters, aiming to

maximize the between group dissimilarity. For

further background details on hierarchical clustering

please see Duda, Hart, and Stork (2004) or Hastie,

Tibshirani and Friedman (2009) which are standard

reference works.

We used hierarchical clustering with Ward’s

linkage to cluster the lower-dimensional

representation obtained following unsupervised

feature selection with iDetect. For further details and

experiments with the full dataset and the lower

dimensional dataset we refer to Tsanas and Arora

(2020). The number of clusters was determined

following visual inspection of the dendrogam as

described in the methodology by Sheaves et al.

(2016).

We used the iDetect algorithm and the

methodology we previously described (Tsanas and

Arora, 2020) to reproduce our findings and use the

same feature subset (21 features, primarily from the

wavelet dysphonia measures) across the three

cohorts. We applied hierarchical clustering

independently for each cohort, using the same feature

subset that has been obtained using iDetect on the

Boston dataset (Tsanas and Arora, 2020). In all cases,

we visualized the dendrograms to visualize the

underlying data structure.

3.5 Data Visualization

We applied the t-distributed Stochastic Neighbor

Embedding (t-SNE) algorithm (van der Maaten and

Hinton, 2008) to obtain a 2D data representation and

visualize the data structure embedded in the high-

dimensional space. We used the 21 features we had

previously identified (Tsanas and Arora, 2020) to

project the 21-dimensional space into 2D. The

resulting representation may provide new insights in

terms of participant assignment in those plots and has

been used to visually annotate the points using the

cluster analysis results.

4 RESULTS

This section is split into two subsections to report on

the generalization of the cluster findings across the

three cohorts, and then to also report on the

generalization of feature selection towards binary

differentiation of PwP and controls.

4.1 Exploring Cluster Generalization

across the Three Cohorts

We applied hierarhical clustering to deterministically

assign cluster membership for each sample.

Subsequently, we applied t-SNE to obtain the 2D data

projection of the feature space spanned by the

selected feature subset, independently for each of the

three cohorts (see Figure 1). We found that across all

three cohorts hierarchical clustering leads to groups

which almost completely agree with the data

projections in 2D space in terms of almost distinct

cluster separation as can be visually affirmed by

Figure 1. This is particularly revealing given that the

data projection and clustering algorithms operate

independently, and these plots serve to intuitively

validate the cluster groupings. We defer further

elaboration for the Discussion.

Assessing Parkinson’s Disease Speech Signal Generalization of Clustering Results across Three Countries: Findings in the Parkinson’s

Voice Initiative Study

127

Figure 1: Two-dimensional representation of the datasets

with selected features using t-SNE and marking of the four

clusters (denoted C1…C4) computed using hierarchical

clustering with the selected feature subset from Tsanas and

Arora (2020).

4.2 Assessing Generalization of Selected

Features for Binary Differentiation

So far we have used data only from the PD

participants in each of the three cohorts, aiming to

derive clusters and assess cluster consistency. As a

final exploratory step, we wanted to apply a

supervised feature weighting algorithm to determine

whether there is also consistency in the key

contributing features to differentiate PwP from

controls across the three cohorts.

We present the results of the LOGO weights in

Fig. 2 for all three cohorts to faciliate visual

comparison. We remark that the actual weights in

LOGO are affected by the number of samples in the

dataset. The primary observation, however, is that

there is again good consistency on the top selected

features across the three datasets. We summarize the

selected features in descending order for each of the

three datasets in Table 2. There is overall agreement

across the datasets on the key contributing features,

and the algorithmic families those features represent.

Figure 2: Feature weights computed using LOGO for each

of the three cohorts in the study.

Table 2: Summary of LOGO-selected features in

descending order for each of the three cohorts.

Boston Oxfor

Toronto

Feature name

VFER

SEO

Jitte

F0-TKEO

rc95

Jitte

F0-TKEO

rc95

MFCC OQ

std

closed

F0 - F0

VFER

TKEO

VFER

TKEO

std

closed

Jitte

F0-TKEO

rc25

VFER

std

Jitte

itch-TKEO

rc25

MFCC Jitte

F0-TKEO

rc95

GNE

SNR

TKEO

std

closed

MFCC 11

MFCC

MFCC 9

MFCC Shimme

TKEO

rc95

MFCC Jitte

itch-TKEO

rc25

det LT entropy

MFCC 12

MFCC 1

det LT entropy

MFCC 6

det LT entropy Shimme

TKEO

rc25

For brevity we only present the top-10 selected features using

LOGO.

BIOSIGNALS 2021 - 14th International Conference on Bio-inspired Systems and Signal Processing

128

5 DISCUSSION

We extended our previous work to assess the

generalization of findings across the three English-

speaking cohorts in PVI. We demonstrated that the

methodology we had previously developed in the

Boston cohort for cluster membership assignment

using the exact parameters we had previously

reported (Tsanas and Arora, 2020), generalizes very

well for the Oxford and Toronto cohorts in PVI.

There is strong internal consistency in identifying

four PwP clusters, which are almost clearly separable

as indicated in Fig. 1 when projecting data into a 2D

transformed feature space. Moreover, we identified

similar features that jointly contribute the

differentiation of PwP and controls (Fig. 2 and Table

2) which further supports the generalizability of those

findings, at least for the English-speaking cohorts.

Similarly to other clinical conditions, there are

important implications and translational potential for

cluster findings. For this particular setting, we

envisage a newly diagnosed PwP could be

phenotyped using sustained vowels to be assigned in

a PD cluster, which could provide information about

symptom trajectory or optimal treatment to follow on

the basis of similarity to other PwP within the same

cluster. It is often possible to provide a tentative

interpretation of clusters using additional

information, e.g. regarding PD symptom trajectory or

targeted symptoms/therapies.

We remark that our findings are strongly

supporting previous studies on PwP subtyping, which

had similarly reported the identification of four

clusters. Indicatively, Lewis et al. (2005), collected

demographic, motor, mood, and cognitive measures

from 120 early-stage PwP and applied standard k-

means resulting into four clusters: (1) younger PD

onset; (2) tremor-dominant; (3) non-tremor dominant

with considerable cognitive impairment and mild

depression; and (4) rapid disease progression but no

cognitive impairment. Similarly, Lawton et al. (2018)

used standardized questionnaires to assess motor,

non-motor, and cognitive domains on two PD cohorts

(1601 and 944 participants). They reported four main

subgroups: (1) fast motor progression with

symmetrical motor disease, poor olfaction, cognition

and postural hypotension; (2) mild motor and non-

motor disease with intermediate motor progression;

(3) severe motor disease, poor psychological well-

being and poor sleep with an intermediate motor

progression; (4) slow motor progression with tremor-

dominant, unilateral disease. van Rooden et al. (2011)

similarly reported four subgroups: (1) mildly affected

in all domains, (2) predominantly severe motor

complications, (3) affected mainly on

nondopaminergic domains with no major motor

complications, (4) severely affected across all

domains. Mu et al. (2017) assessed motor and non-

motor symtoms in two cohorts (411 and 540

participants), and also reported four clusters: (1) mild,

(2) non-motor dominant, (3) motor-dominant, and (4)

severe. We stress that these studies had used different

data modalities, which further serves to underline the

important validity of speech towards providing

holistic information about motor and other PD

symptoms (Tsanas, 2012).

The findings in Fig. 1 make a very compelling

case regarding cluster validation: using

independently cluster analysis and 2D data projection

we find that the computed clusters can be visually

verified. However, it is not directly obvious how well

the four clusters reported herein computed using

acoustic features extracted from sustained vowels

match with the underlying PD symptoms and clusters

of the preceding studies (Lewis et al. 2005; van

Rooden et al., 2011; Lawton et al., 2018).

Unfortunately, in the PVI study we had not collected

additional symptom based entries in the form of

patient reported outcome measures or clinical

assesssments. On the other hand, studies which have

longitudinal clinical evaluations and patient reported

outcome measures do not have speech signal

recordings which would enable to explore bridging

this gap. Applying a range of signal processing and

data analytics tools across different modalities, with

the ultimate aim of fusing information can provide a

more holistic translational path for clinical research

(Gorriz et al., 2020; Woodward et al., 2020).

We emphasize that many clustering studies

focusing on clinical data in general and in PD

research in particular, rely on tools which make rigid

assumptions such as k-means (e.g. Lewis et al., 2005;

Lawton et al., 2018). This technique, although simple

to apply has some fundamental drawbacks (Hastie,

Tibshirani, Friedman, 2009; Duda, Hart, Stork,

2001). Further challenges in cluster analysis include

selecting a robust feature subset which could better

reveal the underlying groups without having any

labels available (Dy and Brodley, 2004),

standardizing variables or introducting weights for

different variables, and validating findings. In

practice, many of these crucial implementation details

in the application of cluster analysis methodology in

often omitted. For an overview of this field,

Assessing Parkinson’s Disease Speech Signal Generalization of Clustering Results across Three Countries: Findings in the Parkinson’s

Voice Initiative Study

129

challenges, and suggestions for best practice when

reporting clustering results we refer to Horne et al.

(2020).

We envisage these robust cluster findings which

appear to generalize very well may contribute

towards improving understanding of the nature of PD

subtypes and hence potentially be translated to inform

therapeutic interventions in clinical practice

(Triantafyllidis and Tsanas, 2019). We are further

exploring the PVI data to investigate differences

across the English-speaking and other cohorts, both

towards understanding differences versus controls

and also internal variability which may inform future

clinical trials.

ACKNOWLEDGEMENTS

We are grateful to Max Little who led the Parkinson’s

Voice Initiative where the data for this study was

collected, and to Ladan Baghai-Ravary for

developing the data collection process using the

Aculab servers. We would like to extend our thanks

to all participants in the PVI study. The study was

made possible through generous funding via an

EPSRC-NCSML award to AT and SA.

REFERENCES

Arora, S. Visanji, N.P., Mestre, T.A., Tsanas, A., Al

Dakheel, A., Connolly, B.S., Gasca-Salas, C., Kern,

D.S., Jain, J., Slow, E.J., Faust-Socher, A., Lang, A.E.

Little, M.A., Marras C. 2018. Investigating voice as a

biomarker for leucine-rich repeat kinase 2-associated

Parkinson’s disease: a pilot study, Journal of

Parkinson’s Disease, Vol. 8(4), pp. 503-510

Arora, S., Baghai-Ravary, L., Tsanas A. 2019. Developing

a large scale population screening tool for the

assessment of Parkinson’s disease using telephone-

quality speech, Journal of Acoustical Society of

America, Vol. 145(5), 2871-2884

Bishop, C.M. 2006. Pattern recognition and machine

learning, Springer

Camacho, A., Harris, J.G. 2008. A sawtooth waveform

inspired pitch estimator for speech and music, Journal

of the Acoustical Society of America, Vol. 124, 1638-

1652

Dorsey, E.R., George, B.P., Leff, B., Willis A.W. 2013. The

coming crisis: obtaining care for the growing burden

of neurodegenerative conditions, Neurology, Vol. 80,

1989-1996

Duda, R.O., Hart, P.E., Stork, D.G. 2001. Pattern

classification, Wiley-interscience, 2nd ed.

Dy, J.G., Brodley, C.E. 2004. Feature selection for

unsupervised learning, Journal of Machine Learning

Research, Vol. 5, 845-889

Fereshtehnejad, S.-M. et al. 2015. New clinical subtypes of

parkinson disease and their longitudinal progression: a

prospective cohort comparison with other

phenotypes. JAMA Neurology, Vol. 72, pp. 863–873

GBD 2016 Parkinson's Disease Collaborators 2018. Global,

regional, and national burden of Parkinson’s disease,

1990–2016: a systematic analysis for the Global Burden

of Disease Study 2016. The Lancet Neurology, Vol. 17,

pp. 939-953

Gomez-Vilda, P., Mykyska, J., Gomez, A., Palacios, D.,

Rodellar, V., Alvarez A. 2019. Characterization of

Parkinson’s disease dysarthria in terms of speech

articulation kinematics, Biomedical Signal Processing

and Control, Vol. 52, 312-320

Gorriz, J.M., Ramirez, J., Ortiz, A., et al. (2020). Artificial

intelligence within the interplay between natural and

artificial computation: advances in data science, trends

and applications, Neurocomputing, Vol. 410, pp. 237-

270

Guyon, I., Gunn, S., Nikravesh, M., Zadeh, L.A. (Eds.)

2006. Feature Extraction: Foundations and

Applications, Springer

Hartelius L., Svensson P. 1994. Speech and swallowing

symptoms associated with parkinson‘s disease and

multiple sclerosis: A survey, Folia Phoniatr. Logop.,

Vol. 46, pp. 9- 17

Hastie, T. Tibshirani, R. Friedman J. 2009. The elements of

statistical learning: data mining, inference, and

prediction, Springer, 2nd ed.

Horne, E., Tibble, H. Sheikh, A., Tsanas A. 2020.

Challenges of clustering multimodal clinical data: a

review of applications in asthma subtyping, JMIR

Medical Informatics, Vol. 8(5), e16452

Lawton, M., Ben-Shlomo, Y., May, M.T., et al. 2018.

Developing and validating Parkinson’s disease

subtypes and their motor and cognitive progression,

Journal of Neurology, Neurosurgery and Psychiatry,

Vol. 89, pp. 1279-1287

Lewis, S.J.G., Foltynie, T., Blackwell, A.D., Robbins,

T.W., Owen, A.m., Barker R.A. 2005. Heterogeneity

of Parkinson’s disease in the early clinical stages using

a data driven approach, Journal of Neurology,

Neurosurgery and Psychiatry, Vol. 76, 343-348

Mu, J., Chaudhuri, K.R., Bielza, C., de Pedro-Cuesta, J.,

Larranaga, P., Martinez-Martin, P. 2017. Parkinson’s

disease subtypes identified from cluster analysis of

motor and non-motor symptoms, Frontiers in Aging

Neuroscience, 9:301

Nguyen, A., Roth, N., Ghassemi, N.H., Hannink, J., Seel,

T., Klucken, J., Gassner, H., Eskofier, B.M. 2019.

Development and clinical validation of inertial sensor-

based gait-clustering methods in Parkinson’s disease,

Journal of NeuroEngineering and Rehabilitation,

16:77

Olanow, C.W., Stern, M.B., Sethi,K. 2009. The scientific

and clinical basis forthe treatment of Parkinson

disease, Neurology, Vol. 72 (21 Suppl 4) s1-s136

Rueda, A., Krishnan, S., 2018. Clustering Parkinson’s and

age-related voice impairment signal features for

unsupervised learning, Advances in Data Science and

Adaptive Analysis, Vol. 10(2);1840007

San Segundo, E., Tsanas, A., Gomez-Vilda, P., 2017.

Euclidean distances as measures of speaker similarity

including identical twin pairs: a forensic investigation

BIOSIGNALS 2021 - 14th International Conference on Bio-inspired Systems and Signal Processing

130

using source and filter voice characteristics, Forensic

Science International, Vol. 270, pp.25-38

Selikhova, M., Williams, D.R., Kempster, P.A., Holton,

J.L., Revesz, T., Lees, A.J. 2009. A clinic-pathological

study of subtypes in Parkinson’s disease, Brain, Vol.

132, pp. 2947-2957

Sheaves, B., Porcheret, K., Tsanas, A., Espie, C., Foster, R.,

Freeman, D., Harrison, P.J., Wulff, K., Goodwin,

G.M. 2016. Insomnia, nightmares, and chronotype as

markers of risk for severe mental illness: results from

a student population, Sleep, Vol. 39(1), pp. 173-181

Sun, Y., Todorovic, S., Goodison, S. 2010. Local-learning-

based feature selection for high-dimensional data

analysis, IEEE Transactions on Pattern Analysis and

Machine Intelligence, Vol. 32(9), pp. 1610-1626

Titze, I.R. 2000. Principles of Voice Production. National

Center for Voice and Speech, Iowa City, US, 2nd

printing

Triantafyllidis, A.K., Tsanas A. 2019. Applications of

machine learning in real-life digital health

interventions: review of the literature, Journal of

Medical Internet Research (JMIR), Vol. 21(4), e12286

Tsanas, A., Little, M.A., McSharry, P.E., Ramig, L.O.

2010a. New nonlinear markers and insights into

speech signal degradation for effective tracking of

Parkinson’s disease symptom severity, International

Symposium on Nonlinear Theory and its

Applications (NOLTA), pp. 457-460, Krakow,

Poland, 5-8 September

Tsanas, A., Little, M.A., McSharry, P.E., Ramig, L.O.

2010b. Enhanced classical dysphonia measures and

sparse regression for telemonitoring of Parkinson's

disease progression, IEEE Signal Processing Society,

International Conference on Acoustics, Speech and

Signal Processing (ICASSP), pp. 594-597, Dallas,

Texas, US, 14-19 March 2010

Tsanas, A., Little, M.A., McSharry, P.E., Ramig, L.O.

2011. Nonlinear speech analysis algorithms mapped to

a standard metric achieve clinically useful

quantification of average Parkinson’s disease

symptom severity, Journal of the Royal Society

Interface, Vol. 8, pp. 842-855

Tsanas, A., Little, M.A., McSharry, P.E., Spielman, J.,

Ramig, L.O. 2012. Novel speech signal processing

algorithms for high-accuracy classification of

Parkinson’s disease, IEEE Transactions on

Biomedical Engineering, Vol. 59, 1264-1271

Tsanas A. 2012. Accurate telemonitoring of Parkinson’s

disease symptom severity using nonlinear speech

signal processing and statistical machine learning,

Ph.D. thesis, Oxford Centre for Industrial and Applied

Mathematics, University of Oxford

Tsanas A. 2013. Acoustic analysis toolkit for biomedical

speech signal processing: concepts and algorithms, 8th

International Workshop on Models and Analysis of

Vocal Emissions for Biomedical Applications

(MAVEBA), pp. 37-40, Florence, Italy, 16-18

December

Tsanas, A., Gómez-Vilda P., 2013. Novel robust decision

support tool assisting early diagnosis of pathological

voices using acoustic analysis of sustained vowels,

Multidisciplinary Conference of Users of Voice,

Speech and Singing (JVHC 13), pp. 3-12, Las Palmas

de Gran Canaria, 27-28 June

Tsanas, A. Little, M.A., McSharry P.E. 2013. A

methodology for the analysis of medical data, in

Handbook of Systems and Complexity in Health, Eds.

J.P. Sturmberg, and C.M. Martin, Springer, pp. 113-

125 (chapter 7)

Tsanas, A., Little, M.A., Fox, C., Ramig L.O. 2014a.

Objective automatic assessment of rehabilitative

speech treatment in Parkinson’s disease, IEEE

Transactions on Neural Systems and Rehabilitation

Engineering, Vol. 22, 181-190

Tsanas, A., Zañartu, M., Little, M.A., Fox, C., Ramig, L.O.,

Clifford, G.D. 2014b. Robust fundamental frequency

estimation in sustained vowels: detailed algorithmic

comparisons and information fusion with adaptive

Kalman filtering, Journal of the Acoustical Society of

America, Vol. 135, 2885-2901

Tsanas A. 2019. New insights into Parkinson’s disease

through statistical analysis of standard clinical scales

quantifying symptom severity, 41st IEEE Engineering

in Medicine in Biology Conference (EMBC), Berlin,

Germany, 23-27 July

Tsanas, A., Arora S. 2019. Biomedical speech signal

insights from a large scale cohort across seven

countries: the Parkinson’s voice initiative study,

Models and Analysis of Vocal Emissions for

Biomedical Applications, Florence, Italy, 17-19

December 2019

Tsanas, A., Arora S. 2020. Large-scale clustering of people

diagnosed with Parkinson’s disease using acoustic

analysis of sustained vowels: findings in the

Parkinson’s voice initiative study, 13th International

Joint Conference on Biomedical Systems and

Technology (BIOSTEC), pp. 369-376, Malta, 26-28

February 2020

Tsanas, A., Little, M.A., Ramig L.O. Remote assessment of

Parkinson’s disease symptom severity using the

simulated cellular mobile telephone network, IEEE

Access (in press)

Woodward, K. Kanjo, E., Brown, D., et al. (2020). Beyond

mobile apps: a survey of technologies for mental well-

being, IEEE Transactions Affective Computing, (in

press)

van der Maaten, L.J.P., Hinton, G.E. 2008. Visualizing

high-dimensional data using t-SNE, Journal of

Machine Learning Research , Vol. 9, pp. 2579-2605

van Rooden S.M., Colas F., Martínez-Martín P., Visser M.,

Verbaan D., Marinus J., Chaudhuri R.K., Kok J.N.,

van Hilten J.J. 2011. Movement Disorders, Vol. 26(1),

pp.51-58

Yao, J., Mao, Q., Goodison, S., Mai, V., Sun, Y. 2015.

Feature selection for unsupervised learning through

local learning. Pattern Recognition Letters, Vol. 53,

pp. 100–107

Zhang, X., Chou, J., Liang, J., Xiao, C., Zhao, Y., Sarva,

H., Henchcliffe, C., Wang, F. 2019. Data-driven

subtyping of Parkinson’s disease using longitudinal

clinical records: a cohort study, Scientific Reports,

Vol. 9;797

Assessing Parkinson’s Disease Speech Signal Generalization of Clustering Results across Three Countries: Findings in the Parkinson’s

Voice Initiative Study

131