The practice of digital pathology not only allows for quantitative slide analysis but also a shift in the way that we think about using slides. Instead of characterizing a slide based upon IHC staining intensity or a few simple features, thousands of features can be extracted instantaneously from large populations of slides. Using machine learning and other advanced classification approaches, we can then mine these features from slides for unique biomarkers. For example, we might determine that the combination of several features such as nuclei shape, size and distribution are indicators of an early stage of a specific type of disease progression. Therefore, the practice of transforming classical glass microscope slides into digital data sets allows us to uncover unique features and biomarkers that a human would never be able to look for.
How we work with Clients
A typical machine learning project starts with a Client sending us physical glass microscope slides for scanning or a hard drive of digital pathology files. These files are typically annotated by the customer into two or more groups such as normal and abnormal samples. The aim of most of these projects is to identify unique features which in a biased or unbiased manner allow us to discriminate these different populations of slides from one another. This approach can then be used with the Client going forward to automtically classify unknown samples.