Title: A computer-aided diagnosis system to detect Parkinson disease by using acoustic features
Authors: Javier Carron - Universidad de Extremadura (Spain)
Yolanda Campos-Roca - Universidad de Extremadura (Spain)
Mario Madruga Escalona - Universidad de Extremadura (Spain)
Carlos Javier Perez Sanchez - University of Extremadura (Spain) [presenting]
Abstract: Parkinson's Disease (PD) is a chronic progressive neurodegenerative disorder that has an impact on the patients' voice, among other symptoms. Early detection is key to improving the patients' quality of life. A computer-aided diagnosis system to detect PD has been developed. This system collects voice recordings from an app, extracts multiple relevant acoustic features, and uses a two-stage variable selection and classification method that has been specifically designed for this task. An experiment has been conducted to analyze the system performance. A total of 30 people affected by PD were recruited among voluntary members of the Regional Association for Parkinson's Disease of Extremadura (Spain). Also, 30 healthy subjects were recruited to approximately match age and sex. After obtaining the voice recordings, a feature extraction process is performed to provide 33 acoustic features. The numerical features fed 5 variable selection and classification methods, and the selected features and performance metrics were compared. The best accuracy rate of 92\% was obtained under a stratified cross-validation framework. The same methodology was applied to a sex and age-matched subsample from an existing voice recording database (https://parkinsonmpower.org), leading to a considerably lower best accuracy rate of 71\%. The results clearly show the importance of the voice recordings and the potential of the proposed system.