Assessing Parkinson’s Disease at Scale Using Telephone-Recorded Speech: Insights from the Parkinson’s Voice Initiative

Abstract

Numerous studies have reported on the high accuracy of using voice tasks for the remote detection and monitoring of Parkinson’s Disease (PD). Most of these studies, however, report findings on a small number of voice recordings, often collected under acoustically controlled conditions, and therefore cannot scale at large without specialized equipment. In this study, we aimed to evaluate the potential of using voice as a population-based PD screening tool in resource-constrained settings. Using the standard telephone network, we processed 11,942 sustained vowel /a/ phonations from a US-English cohort comprising 1078 PD and 5453 control participants. We characterized each phonation using 304 dysphonia measures to quantify a range of vocal impairments. Given that this is a highly unbalanced problem, we used the following strategy: we selected a balanced subset (n = 3000 samples) for training and testing using 10-fold cross-validation (CV), and the remaining (unbalanced held-out dataset, n = 8942) samples for further model validation. Using robust feature selection methods we selected 27 dysphonia measures to present into a radial-basis-function support vector machine and demonstrated differentiation of PD participants from controls with 67.43% sensitivity and 67.25% specificity. These findings could help pave the way forward toward the development of an inexpensive, remote, and reliable diagnostic support tool for PD using voice as a digital biomarker.

Data-Driven Subtyping of Parkinson’s Using Acoustic Analysis of Sustained Vowels and Cluster Analysis: Findings in the Parkinson’s Voice Initiative Study

“…The success of the wavelet-based dysphonia measures in previous PD-related applications [18,20,22] which was also verified in this study is particularly noteworthy over other dysphonia measures which have often dominated the speech signal processing field such as jitter and shimmer [32]. In accordance with results we had reported in related PVI studies in discriminating PwP from controls [30,34], it appears that the nonlinear dysphonia measures (which were very successful in applications when sustained vowels were collected under carefully controlled acoustic conditions, e.g. to replicate UPDRS [18][19][20]), are not particularly useful towards PD subtyping.…”

“…Moreover, we remark that all analysis in this study towards determining PD subtypes relied solely on sustained vowel /a/ phonations (the only speech modality available in PVI). Although there is a considerable body of research work to support the use of sustained vowels in different PD applications [19,21,25,33,34,62], it is conceivable that the use of additional speech modalities (e.g. running speech) might provide additional complementary information towards informing PD subtypes.…”

Comparing Lab-based and Telephone-based Speech Recordings Towards Parkinson's Assessment: Insights from Acoustic Analysis

“…Whilst this is the right first step to demonstrate feasibility under favorable conditions and minimize heterogeneity, current findings may be challenging to scale up in practice if there is a strict requirement of lab-based conditions. To enable large scale analysis into PD we set up a large international multi-site trial, the Parkinson's Voice Initiative (PVI), collecting more than 19,000 sustained vowel /a/ recordings over the standard telephone network, with PwP across 7 countries [16]- [18]. Although the data collected in PVI is not of the same high quality as data collected under carefully controlled acoustic conditions in the lab, the large number of samples facilitates new explorations.…”

“…Although the data collected in PVI is not of the same high quality as data collected under carefully controlled acoustic conditions in the lab, the large number of samples facilitates new explorations. However, the reduced data quality poses new challenges, and current algorithms have led to considerable performance degradation in the binary differentiation of PwP and controls [16], [18] compared to the very promising results using the exact data processing methodology we have previously reported when processing lab-based data [5].…”

“…Please cite this as: A. The study uses two databases: the Intel At-Home Testing Device (AHTD) [4], [6] and the PVI [16]- [18]. The former used a high-quality device where 42 PwP in the US collected data at home weekly for six months, giving rise to 5875 sustained vowel /a/ phonations.…”