Discrimination of Parkinson's disease participants from healthy controls using telephone-quality voice recordings

S. Arora, A. Tsanas

Research output: Contribution to journalMeeting abstractpeer-review

Abstract / Description of output

Objective: To test the practicality and efficacy of telephone-quality voice recordings to discriminate Parkinson’s disease (PD) participants from healthy controls (HC).

Background: Vocal performance degradation is met in the vast majority of people diagnosed with PD, and may be one of the earliest indicators of disease onset. Using high-quality voice recordings, recent studies have developed technologies both to discriminate PD from HC, and also for symptom severity telemonitoring. However, these studies may be limited in scaling massively across the population as they rely on expensive specialized equipment to collect the data, which might not be available in resource-constrained settings. In this study, we investigated whether telephone-quality voice recordings collected using readily available standard commercial consumer phones could be used to provide easily accessible, cost-effective means towards accurate PD assessment.

Methods: We collected sustained vowel phonations (‘aaah’ sounds, where the subject is requested to keep the pitch and amplitude as steady as possible) through telephone-quality digital audio lines, under realistic, non-lab conditions. The recordings were obtained from the following locations: Argentina, Brazil, Canada, Spain, Mexico, UK and USA. After screening out bad recordings, we used 2799 recordings from 1507 PD participants (mean age: 50.9 years, 45.9% female), and 15486 recordings from 8394 control participants (mean age: 51.1 years, 45.0% female). We extracted and analyzed 309 dysphonia measures to quantify subtle changes in the recorded signals, which can differentiate PD from HC voices. We presented the dysphonia measures to a random forest classifier, and assessed the efficacy of the model using 10-fold cross-validation with 100 iterations for statistical confidence.

Results: Using only the voice recordings, we achieved a mean out of sample sensitivity of 63.8% (standard deviation 2.3%) and mean specificity of 67.2% (standard deviation 2.9%) in discriminating PD from HC.

Conclusions: The discrimination results obtained are promising and are considerably better than standard naïve benchmarks (or chance). These findings warrant further investigation into the feasibility of using voice as a potential biomarker of PD, even with low-quality telephone-based recordings.
Original languageEnglish
Pages (from-to)S266-S266
Number of pages1
JournalMovement Disorders
Volume31
Publication statusPublished - 21 Jun 2016
Event20th International Congress of Parkinson's Disease and Movement Disorders - Berlin, Germany
Duration: 19 Jun 201623 Jun 2016

Fingerprint

Dive into the research topics of 'Discrimination of Parkinson's disease participants from healthy controls using telephone-quality voice recordings'. Together they form a unique fingerprint.

Cite this