Developing a large scale population screening tool for the assessment of Parkinson's disease using telephone-quality voice

Abstract

Recent studies have demonstrated that analysis of laboratory-quality voice recordings can be used to accurately differentiate people diagnosed with Parkinson's disease (PD) from healthy controls (HC). These findings could help facilitate the development of remote screening and monitoring tools for PD. In this study, we analyzed 2759 telephone-quality voice recordings from 1483 PD and 15321 recordings from 8300 HC participants. To account for variations in phonetic backgrounds, we acquired data from seven countries. We developed a statistical framework for analyzing voice, whereby we computed 307 dysphonia measures that quantify different properties of voice impairment, such as, breathiness, roughness, monopitch, hoarse voice quality, and exaggerated vocal tremor. We used feature selection algorithms to identify robust parsimonious feature subsets, which were used in combination with a Random Forests (RF) classifier to accurately distinguish PD from HC. The best 10-fold cross-validation performance was obtained using Gram-Schmidt Orthogonalization (GSO) and RF, leading to mean sensitivity of 64.90% (standard deviation, SD 2.90%) and mean specificity of 67.96% (SD 2.90%). This large-scale study is a step forward towards assessing the development of a reliable, cost-effective and practical clinical decision support tool for screening the population at large for PD using telephone-quality voice.

New insights into Parkinson’s disease through statistical analysis of standard clinical scales quantifying symptom severity

“…to assess PD symptom severity [5] and for early PD biomarkers [19], thus helping scale up research work when the entire UPDRS is not fully administered in standard clinical practice. This is very well aligned with our recent work aiming to develop tools that can be deployed at large scale in community studies [20].…”

Telemonitoring Parkinson’s disease using machine learning by combining tremor and voice analysis

“…Later, many researchers as like in [26][27][28] worked on different voice data sets to improve the detection accuracy by feature engineering or applying different classifiers for developing telemonitoring systems. They also investigated different feature selection techniques to reduce dimensionality.…”

Remote Assessment of Parkinson’s Disease Symptom Severity Using the Simulated Cellular Mobile Telephone Network

“…For example, there has been consistent interest in the binary differentiation of PD from healthy controls [48] , [66] , [67] amongst different research teams, which has generally led to very successful outcomes. We have recently reported on our findings in the Parkinson’s Voice Initiative (PVI), a large international study where we had collected more than 19,000 sustained vowel /a/ phonations across seven countries [30] . The PVI phonations were collected under acoustically non-controlled conditions over the phone with the explicit aim of investigating large scale population screening towards PD assessment using telephone-quality speech.…”

“…We had demonstrated clinically meaningful differentiation of PWP versus controls, thus highlighting the potential of this technology at scale. Recently, researchers have started exploring speech data from different corpora which may lead to new insights across PD populations with different linguistic backgrounds [30] , [68] . Furthermore, some studies have investigated different speech tasks for PD evaluations [60] , [68] , although it is still early to decide whether any particular task is clearly better that competing approaches.…”

“…Thus, the standard phone network provides convenient means towards inexpensive and frequent PD severity assessments, facilitating monitoring and potentially assisting rehabilitation. Data-mining of speech signals obtained using the public telephone network to extract clinically useful information has recently shown promising results [28] – [30] . Similarly, Saenz-Lechon et al [31] investigated the effect of different data transmission rates in automatic voice pathology detection, and concluded that compressing signals (down to at most 64 kbps) does not prevent accurate detection of vocal pathologies.…”