Achieving Clinical Utility for 2D Digital Pathology

Achieving Clinical Utility for 2D Digital Pathology
Contact for more information: 

John A. Foekens

Breast cancer is not just one disease. Several molecular subtypes of breast cancer have been recognized that are associated with differential disease outcome as well as differential treatment response. Consequently, patients should receive personalized treatment based on the molecular features of their tumors. We use 2D digital pathology in order to measure and quantitate these specific features that facilitate diagnostic pathologists with the appropriate tools for clinical decision making.  

Gene Expression Profiling
Achieving Clinical Utility for 2D Digital Pathology

A gene expression profile for a tumor represents the activity state of all human genes, the so-called global mRNA expression profile, of that tumor. Gene expression profiles vary among different breast tumors. Grouping breast tumors with similar gene expression profiles has distinguished the different molecular subtypes of breast cancer: luminal, basal-like and ERBB2+ subtypes. However, gene expression profiles may also be used to extract gene expression differences between predefined groups of tumors. This way we have unraveled the genes that are differentially expressed between patients that do not need treatment after surgical resection of their primary breast tumor and patients that do. These genes were identified for luminal, basal-like and ERBB2+ subtypes specifically. Currently, almost all breast cancer patients are treated with some form of additional systemic therapy after surgical resection of the tumor. Identifying which patients do not benefit from this therapy would prevent unnecessary treatment toxicity for these patients, but also lead to a reduction in healthcare costs.

Translation to Protein Markers
Achieving Clinical Utility for 2D Digital Pathology

Gene expression profiles are not easily implemented in routine diagnostic pathology practice. In fact, immunohistochemical assays, in which a protein is visualized by binding of an antibody with a chromogen attached to it, is the preferred method of choice. Therefore, we developed immunohistochemical assays for a particular selection of the differentially expressed prognostic genes. This selection was based on several criteria, but most importantly it captured the prognostic value as identified by gene expression profiling.

The image shows an immunohistochemical staining in which the brown color indicates the presence of the protein of interest, which is located on the cell membrane.

Algorithmic Detection of Protein Staining
Achieving Clinical Utility for 2D Digital Pathology

As a final step, the protein staining in the immunohistochemical analyses needs to be measured and quantified. For this, a 2D digital pathology scanner scans the slides in order to make high resolution images. Machine learning methods are deployed in order to develop algorithms that can accurately quantify the protein staining in a predefined area of the tumor. This way, the 2D digital pathology platform, including scoring algorithms, are able to assist the pathologist in providing accurate measurements to back up clinical decision making.

The image shows the same immunohistochemical staining as above, but with an overlay visualizing positive staining as indicated by the scoring algorithm.