Image Informatics for the Characterization of Molecular Subtypes in Breast Carcinoma Tissue 7a.019.DU

Project Start Date: Aug 1, 2018
Research Areas: Analytics, Analytics - Machine Learning
Funding: Member Funded
Project Tags: ,

Project Summary

Histological examination of tumor and biopsy specimens remains the key diagnostic tool for pathology diagnosis and staging. The availability of large-scale architectural information and fine-scale features can serve as important cues from which to judge the aggressiveness of the tumor and the patient’s prognosis. The predictive capabilities of histological image analysis, enhanced by informatics techniques, may be harnessed to objectively and reproducibly distinguish tumor subtypes. The key advantage of this approach tackles the two major criticisms of molecular subtyping: 1) the lack of spatial information, that makes gene expression analysis susceptible to artifacts in the presence of tumor heterogeneity, can be overcome with image analysis; 2) by defining tumor molecular subtype morphologically with a reduced number of variables (on the order of tens, rather than thousands), the “curse of dimensionality” no longer places a constraint on our ability to define groups based on pattern analysis.