Computational Image Analysis To Identify a Morphological Signature of Claudin-Low Breast Cancer
Kamaljeet Singh, Laura Collins, Nicholas Knoblauch, Marco Hefti, Andrew Beck. Beth Israel Deaconess Medical Center, Boston, MA
Background: The Claudin-low molecular subtype of breast cancer has recently been identified as a distinct subset of triple negative breast cancers that tends to show decreased expression of the Claudin genes and metaplastic morphologic features, suggestive of an epithelial-mesenchymal transition (EMT) phenotype. Our study aims to further characterize morphologic differences between Basal-like and Claudin-low breast cancers.
Design: To characterize the morphologic signature of this breast cancer subset, we performed an adaption of the C-Path (Computational Pathologist) method for whole-slide images. In brief, the C-Path method segments a histologic image into 3 regions (epithelium, stroma, and an indeterminate region termed “EMT-like”). C-Path then computes thousands of quantitative morphologic characteristics from these regions, characterizing the size, shape, texture, intensity, variability, and relationships between objects within the tissue regions. C-Path was performed on a total of 57 breast cancers that had undergone molecular profiling and subtyping as part of The Cancer Genome Atlas Project and had been classified as Basal-like (n=52) or Claudin-low (n=5). We performed an unpaired two-class Significance Analysis of Microarrays to identify morphologic features significantly associated with the Claudin-low subtype. We set a significance threshold of an FDR < 15% for identifying significantly differentially expressed morphologic features.
Results: We identified 95 morphologic features significantly associated with the Claudin-low molecular subtype at the FDR threshold of 15% and 2 features at a threshold of <1%. 73 (77%) of these features are morphologic characteristics of the breast cancer stroma (including: variability of stromal matrix intensity, with less variability in Claudin-low), 20 (21%) of the features are of the EMT-like tissue (including: textural correlation, with higher values seen in the Claudin-low cancers), and only 2 (2%) are morphologic characteristics of the epithelium (both features are measures of the darkest pixel intensity in a region, with higher values observed in Claudin-low, potentially reflecting the influence of intra-epithelial inflammatory cells).
Conclusions: In this broad analysis of quantitative morphologic differences between Claudin-low and Basal-like breast cancer, we identified morphologic features significantly associated with the Claudin-low molecular subtype. The differentially expressed features are almost exclusively from the stroma and EMT-like tissue with few significant differences identified in the epithelial compartment.
Monday, March 4, 2013 9:30 AM
Poster Session I Stowell-Orbison/Surgical Pathology/Autopsy Awards Poster Session # 40, Monday Morning