Many researchers in the preclinical space who employ histological staining methods are interested in getting more information from their image data and oftentimes traditional pathological evaluation of tissue sections is limited in terms of quantifiable endpoints and throughput. For this reason, many folks are looking to more quantitative computer-based methods of evaluation of image data which are referred to broadly as digital pathology. The tools available in this space can range from both traditional image processing software and algorithms to more advanced methodology like using machine learning for image classification that utilizes extensive training sets, which only recently became available because of the adoption of whole slide imagers (WSI’s) and digitization of histology slides.
The range of quantifiable outputs that can be measured with digital pathology techniques can be simple, such as total cell counts or nuclear morphology, or more complex, such as colocalization and level of expression for two biomarkers of interest. One of the most common applications of digital pathology in evaluation of whole slide images of tissue sections is in the examination of the tumor microenvironment and characterizing different sub-populations of cells, which is often relevant to immune-oncology research. For example, phenotyping and quantifying immune cells such as activated T-cells or resident macrophages near the periphery and within tumor margins is paramount to understanding the effects a particular type of immunotherapy may have. The advantage of using this approach here over more traditional methods of cell-counting such as flow cytometry is the associated spatial information that would be relevant to making meaningful conclusions about the data is not lost.
Other applications of digital pathology used in the quantification of image data is in identifying and measuring tumor area/extent of invasion, for segmenting blood vessels and quantifying angiogenesis, identifying and quantifying tumor cells expressing diagnostically significant biomarkers (ER/PR/HER2 for breast cancer), and for assessing tumor heterogeneity. The ability to extract meaningful quantitative insights from images extends far beyond what can be ascertained with visual examination, and digital pathology is leading the way to make sure that those data are obtainable.
Figure 1. Example of tumor area fraction in breast cancer lumpectomy tissue biopsy H&E image, determined using a machine learning approach.