Automated recognition of cellular phenotypes by support vector machines with feature reduction

  • Authors:

  • Z. Xia;X. Zhou;Z. Yin;Y. Sun;S. T. C. Wong;Y. Mao

  • Affiliations:

  • Zhejiang University, National Laboratory of Industrial Control Technology, Institute of Industrial Process Control, Hangzhou 310027, P.R. China;Harvard University, Harvard Center for Neurodegeneration and Repair, Harvard Medical School and Brigham and Women's Hospital, Harvard Medical School, 220 Longwood Avenue, Goldenson Building 524, B ...;Zhejiang University, National Laboratory of Industrial Control Technology, Institute of Industrial Process Control, Hangzhou 310027, P.R. China;(Correspd. Tel.: +86 0571 87952010/ E-mail: yxsun@iipc.zju.edu.cn) Zhejiang University, National Laboratory of Industrial Control Technology, Institute of Industrial Process Control, Hangzhou 3100 ...;Harvard University, Harvard Center for Neurodegeneration and Repair, Harvard Medical School and Brigham and Women's Hospital, Harvard Medical School, 220 Longwood Avenue, Goldenson Building 524, B ...;Zhejiang University, National Laboratory of Industrial Control Technology, Institute of Industrial Process Control, Hangzhou 310027, P.R. China

  • Venue:
  • International Journal of Knowledge-based and Intelligent Engineering Systems - Extended papers selected from KES-2006
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

In this paper, wrapper based feature selection by support vector machine is used for cellular multi-phenotypic mitotic analysis (MMA) in high content screening (HCS). Haralick texture feature subset and Zernike polynomial moment subset are used respectively or combined together as extracted digital feature set for original cellular images. Feature reduction is done by support vector machine based recursive feature elimination algorithm on these feature sets. With optimal feature subset selected, fuzzy support vector machine are adopted to judge the cellular phenotype. The results indicate Haralick texture feature subset is complementary with Zernike polynomial moment subset, when these two feature subsets are combined together; the cellular phase identification system achieved 99.17% accuracy, which is better than only one feature subset of them is used. The recognition accuracy with feature reduction is better than that achieved when no feature reduction done or using PCA as feature recombination tool on these datasets.