Voice Biomarker Identification for Effects of Deep-Brain Stimulation on Parkinson's Disease
Authors:
Huy Phi,
Sanjeev Janarthanan,
Larry Zhang,
Reza Hosseini Ghomi
Abstract:
Deep-Brain Stimulation (DBS) is a therapy used in conjunction with medication to help alleviate the motor symptoms of Parkinson's Disease (PD). However, the monitoring and adjustment of DBS settings is tedious and expensive, requiring long programming appointments every few months. We investigated the possible correlation between PD motor score severity and digitally extracted patient voice featur…
▽ More
Deep-Brain Stimulation (DBS) is a therapy used in conjunction with medication to help alleviate the motor symptoms of Parkinson's Disease (PD). However, the monitoring and adjustment of DBS settings is tedious and expensive, requiring long programming appointments every few months. We investigated the possible correlation between PD motor score severity and digitally extracted patient voice features to potentially aid clinicians in their monitoring and treatment of PD with DBS, and eventually enable a closed-loop DBS system. 5 DBS PD patients were enrolled. Voice samples were collected for various voice tasks (single phoneme vocalization, free speech task, sentence reading task, counting backward task, categorical fluency task) for DBS ON and OFF states. Motor scores per the Unified Parkinson's Disease Rating Scale (UPDRS) were also collected for DBS ON and OFF states. Voice samples were then analyzed to extract voice features using publicly available voice feature library sets, and statistically compared for DBS ON and OFF. Of the feature categories explored (Acoustic, Prosodic, Linguistic) 6 features from the GeMAPS feature set for acoustic features demonstrated significant differences with DBS ON and OFF (p<0.05). Prosodic features such as pause length/percentage were found to be negatively correlated with increased motor symptom severity. Non-significant differences were found for linguistic features. These findings provide preliminary evidence for acoustic and prosodic speech features to act as potential biomarkers for PD disease severity in DBS patients. We hope to explore further by expanding our data set, identifying other features, applying machine learning models, and working towards a closed-loop DBS system that can auto-tune itself based on changes in a patient's voice.
△ Less
Submitted 25 November, 2019;
originally announced December 2019.
