Model selection for advanced statistical analysis of multiparametric MRI

Advanced quantitative multiparametric MRI techniques allow the imaging of tumor heterogeneity that was before only accessible by histology and the mapping of functional features not available to pathologists. This wealth of information is however difficult to interpret, even by experts in the field, because adequate analysis tools are missing. The combination of quantitative multiparametric MRI, which can assess the tumor heterogeneity, and advanced statistical analysis, is a new concept to improve the diagnostic and the therapeutic orientation of brain tumors. Each pixel can be associated with a “spectrum” which integrates information of each quantitative image. These spectra can then be clustered using statistical methods into a number of clusters. The choice of the number of clusters is an important issue in that the clusters can be matched with histological information and used to define a tumor signature and thus allow diagnosis. In this work we will focus on determining the optimal number of clusters in a data-driven and parsimonious way using advanced Bayesian models and inference methods.

Faculty Supervisor:

Russell Steele

Student:

Partner:

Inria Grenoble - Rhône-Alpes Research Centre

Discipline:

Mathematics

Sector:

University:

McGill University

Program:

Globalink Research Award