Speech analysis for the differential diagnosis between Parkinson's disease, progressive supranuclear palsy and multiple system atrophy - Archive ouverte HAL Access content directly
Master Thesis Year : 2017

Speech analysis for the differential diagnosis between Parkinson's disease, progressive supranuclear palsy and multiple system atrophy

(1)
1

Abstract

Acoustic speech analysis has been shown to have a good potential in differentiation between Parkinson's disease and atypical Parkinsonian syndromes (APS) such as progressive supranuclear palsy (PSP) and multiple system atrophy (MSA). Objective speech features were able to discriminate between PD and APS with 95% accuracy and between PSP and MSA with 75% accuracy in [7]. However, accuracy between PSP and MSA still has a large space to be improved and the more important aim is to provide more explicit information for differential diagnosis. In [7], 75% accuracy was achieved using support vector machine classifier based on radial basis function kernel. This means it's dffcult to interpret the relation between selected features and decision hyperplane. In this internship, for discrimination between PSP and MSA, 9% higher accuracy (i.e, 84% accuracy) was attained by using support vector machine classifier based on radial basis function kernel and 80% accuracy was attained using linear dimension reduction methods and linear classifier. More importantly, with this strategy, we obtain a better understanding of feature discriminative power. This can be indeed very useful in clinical application.
Fichier principal
Vignette du fichier
Rapport-Li.pdf (806.55 Ko) Télécharger le fichier
Origin : Files produced by the author(s)
Loading...

Dates and versions

hal-01627868 , version 1 (02-11-2017)

Identifiers

  • HAL Id : hal-01627868 , version 1

Cite

Gongfeng Li. Speech analysis for the differential diagnosis between Parkinson's disease, progressive supranuclear palsy and multiple system atrophy. Signal and Image Processing. 2017. ⟨hal-01627868⟩

Collections

INRIA INRIA2
101 View
423 Download

Share

Gmail Facebook Twitter LinkedIn More