Online Program Home
  My Program

Abstract Details

Activity Number: 161 - SPEED: Nonparametrics and Imaging
Type: Contributed
Date/Time: Monday, July 31, 2017 : 10:30 AM to 12:20 PM
Sponsor: Section on Statistics in Imaging
Abstract #322481 View Presentation
Title: Bayesian Methods for Image Texture Analysis with Applications to Cancer Radiomics
Author(s): Xiao Li* and Michele Guindani and Chaan Ng and Brian P. Hobbs
Companies: University of Texas, School of Public Health, Department of Biostatistics and University of California, Irvine and The University of Texas MD Anderson C and The University of Texas MD Anderson Cancer Center
Keywords: Image Texture Analysis ; spatial gaussian process ; cancer radiomics ; grey level co-occurrence matrix ; machine learning ; classification

Radiomics, as an emerging field in quantitative imaging, encompasses a broad class of analytical techniques. GLCM, as one type of texture features, is a matrix defined over an image to be the distribution of co-occurring gray-level pixels at a given offset and angle. GLCM-based texture features have been used to quantitatively describe tumor phenotypes. Recent literature have interrogated associations with clinical/pathology information with Machine Learning algorithms using GLCM-based texture features. Reducing the multivariate functional structure to a set of summary statistics is potentially reductive, however, masking patterns that describe disease pathogenesis. In this article, we present a Bayesian probabilistic method for count data observed over a lattice with symmetric structure. The approach was utilized to model the entire GLCM as a multivariate response surface, and applied in a cancer detection context to discriminant malignant from benign adrenal lesions using GLCMs arising from pixel-level image data. In case study and simulation study, our proposed method improved classification accuracy compared to current approaches utilizing GLCM-based texture features.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2017 program

Copyright © American Statistical Association