BH
Bryan Ho
Author with expertise in Diagnosis and Treatment of Voice Disorders
Achievements
This user has not unlocked any achievements yet.
Key Stats
Upvotes received:
0
Publications:
1
(0% Open Access)
Cited by:
0
h-index:
9
/
i10-index:
9
Reputation
Biology
< 1%
Chemistry
< 1%
Economics
< 1%
Show more
How is this calculated?
Publications
0

Speech-based identification of L-DOPA ON/OFF state in Parkinson's Disease subjects

Raquel Norel et al.Sep 18, 2018
+2
J
C
R
Background: Parkinson's disease patients (PDP) are evaluated using the unified Parkinson's disease rating scale (UPDRS) to follow the longitudinal course of the disease. UPDRS evaluation is performed by a neurologist, and hence its use is limited in the evaluation of short-term (daily) fluctuations. Subjects taking L-DOPA as part of treatment to reduce symptoms exhibit motor fluctuations as a common complication. Objectives: The aim of the study is to assess the use of speech analysis as a proxy to continuously monitor PDP medication state. Methods: We combine acoustic, prosody, and semantic features to characterize three speech tasks (picture description, reverse counting and diadochokinetic rate) of 25 PDP evaluated under different medication states: ON and OFF L-DOPA. Results: Classification of medication states using features extracted from audio recordings results in cross-validated accuracy rates of 0.88, 0.84 and 0.71 for the picture description, reverse counting and diadochokinetic rate tasks, respectively. When adding feature selection and semantic features, the accuracy rates increase to 1.00, 0.96 and 0.83 respectively; thus reaching very high classification accuracy on 3 different tasks. Conclusions: We show that speech-based features are highly predictive