Texture analysis of transcranial sonographic images for Parkinson disease diagnostics

Authors

  • A. Sakalauskas Kaunas University of Technology
  • A. Lukoševičius Kaunas University of Technology
  • K. Laučkaitė Lithuanian University of Health Sciences

DOI:

https://doi.org/10.5755/j01.u.66.3.656

Keywords:

texture analysis, multiple feature extraction, sequential feature selection, support vector machines, classification rate

Abstract

Parkinson disease (PD) diagnostics based on transcranial sonography (TCS) images are quite subjective and dependent on competence of physician due to low quality of acquired images. The aim of this study is to evaluate potentiality of textural analysis of TCS images for separation between images obtained from healthy and PD affected people. Closer to scanning probe half of midbrain was manually selected as region of interest (ROI) for texture analysis in TCS image. Four hundred thirteen texture features of ROI (110 subjects) were calculated and used for classification during this study. Sequential feature selection method was applied in order to find optimal subset of features for such classification task. Linear support vector machine (SVM) classifier was used for determination of the optimal number of features for new subset.

The classification rate 78.18% was obtained using the calculated texture features. It was concluded that discrimination power of these texture features is directly dependent on the image quality.

http://dx.doi.org/10.5755/j01.u.66.3.656

Downloads

Published

2011-09-23

Issue

Vol. 66 No. 3 (2011)

Section

APPLICATIONS IN BIOLOGY AND MEDICINE

License

Copyright terms are indicated in the Republic of Lithuania Law on Copyright and Related Rights; Articles 4-37.
In the event the above Work is not accepted and is not published in the Publication or is withdrawn by Authors(s) before acceptance by the Publication; this guarantee form becomes null and void; and the submitted Work is not returned to the Author(s).