Bayesian Spatial and Nonparametric Models for Cancer Radiomics
Abstract
The emerging field of cancer radiomics endeavors to characterize intrinsic patterns of tumor phenotypes and surrogate markers of response from medical imaging modalities. Of particular interest, interrogations of “textural” features based on analysis gray-level spatial dependence through the Gray-Level Co-occurrence Matrix (GLCM) have been the focus of several higher profile cancers radiomics articles. Analytical techniques as currently applied which reduces the multivariate functional structure inherent to GLCM to sets of summary statistics for subsequent analysis through the application of regression and machine learning algorithms, have failed to model the GLCM as a multivariate object, however, and often fail to elucidate the predictive power. In this dissertation research, I proposed a novel supervise learning as well as unsupervised learning approach in a Bayesian hierarchical modeling frame work. For supervised learning approach, I present a Bayesian probabilistic modeling framework for the GLCM as a multivariate object as well as describes its application within a cancer detection context based on computed tomography. The methodology, which circumvents processing steps and avoids evaluations of reductive and highly correlated feature sets, uses latent Gaussian Markov random field structure to characterize spatial dependencies among GLCM cells and facilitates classification via predictive probability. For unsupervised learning approach, I develop a Bayesian multivariate probabilistic framework for GLCMs wherein rounded kernel technique is employed to link the observed multivariate counts data to the latent underlying continuous process. The latent spatial association characterizing conditional independence under Gaussian graphs is introduced via a non-parametric Bayesian approach, with base measure depicted by the Gaussian Markov random field. This approach facilitates to capture the cancer subgroup leading to a natural clustering of subjects with similar GLCM patterns through sharing of information.
Subject Area
Biostatistics
Recommended Citation
Li, Xiao, "Bayesian Spatial and Nonparametric Models for Cancer Radiomics" (2018). Texas Medical Center Dissertations (via ProQuest). AAI10747198.
https://digitalcommons.library.tmc.edu/dissertations/AAI10747198