TITLE: Statistical Learning in Application to the Recognition of the Lung Tissue Texture Patterns in HRCT Images

PRESENTER: Alena Shamsheyeva, http://www.cse.unsw.edu.au/db/staff/info/alenas.html, alenas@cse.unsw.edu.au

AFFILIATION:CSE - UNSW, http://www.cse.unsw.edu.au

DATE: Friday 12th August 2005

TIME: 12:00:00

PLACE: CSE Seminar Room K17_113

ABSTRACT:

A method of pattern detection in High-resolution Computed Tomography
(HRCT) images of the lung is presented. The method is based on
combination of statistical machine learning and image processing
techniques. We use the Support Vector Machines (SVM) algorithm for
pixel-wise classification. Pixels are characterized by textural features
obtained from the wavelet transform of an image. Subsequently, the
pixel-wise classification results are post-processed by application of
image processing techniques. The problem of model selection for the SVM
is studied. We compare model selection methods based on estimation of a
generalization error with a test set and an analytical bound. The
analysis of dependence of the model on a size of a training set is
performed. The validity of application of a model tuned on a small
training set to a final SVM classifier training on a large dataset is
investigated. A comparison between isotropic and anisotropic Gaussian
kernel is performed. We show the correspondence of the model selected
for the anisotropic Gaussian kernel to the textural characteristics of a
lung pattern.

BIOGRAPHY OF SPEAKER:

Alena Shamsheyeva is PhD student at CSE. Alena is going to give a
presentaion of her PhD thesis reasearch.

Host:

Seminar Convenor:

Van Hai Ho