Digital pathology is changing the practice of pathology. The initial benefits cited for digital pathology include educational and archival use cases. Additionally, creating digital representations of glass slides transforms the morphologic features contained on the physical glass slides into digital data. This allows for easier collaboration and increased access to the pathology material. All of these factors represent improvements to the traditional tasks that are performed by pathologists in generating diagnostic reports for our patients. The conversion of glass slides (analogue data) to whole slide images (digital data) represent a new opportunity to provide enhanced diagnostic reports. Quantitative image analysis represents one of the new opportunities especially given the recent development of immune based therapy for malignancies.
Prior efforts for quantitative image analysis have principally involved breast cancer – specifically quantification of oestrogen receptor, progesterone receptor and HER2 expression. This image analysis has been limited by technical restrictions. Previously, only small fields could be analysed. This limitation required that a pathologist select a region to be analysed. In this workflow, a mundane task of selecting a field is dependent on the pathologist. Although selecting the field is a minor task, the pathologist represents a bottleneck in efficiently getting the tissue analysed. An additional by-product of having the pathologist select a field is that the pathologist can just simply interpret the results manually. If the digital assessment does not provide substantial improvement over the manual assessment, it is very difficult to push adoption. Oestrogen and progesterone receptors are both nuclear immunostains that represent relatively simplistic stains to interpret, however, HER2 is a membranous protein and the intensity of the immunostain is critical in the interpretation. Additionally, breast carcinoma is often intermingled with benign tissue elements that must be separated from the carcinoma for report purposes. This intermingling causes difficulty with tumour selection and impedes the use of digital quantification. However, increased billing revenue has been attributed to the digital quantification and so some institutions have begun using digital quantification for reporting out results.
Proliferative index of malignancies can be assessed by Ki-67 immunostain – neuroendocrine neoplasms represent the most common tumour type that utilises this for prognosis. Although this is a nuclear stain, it is still largely reported out by “eyeball” method. The reporting interpretations do not require a fine assessment beyond “eyeball” method and so this has become the practice. Within the last few years, technical advances have been made that allow for whole slide image analysis to be performed. Additionally, artificial intelligence algorithms are being developed that may minimise the need for a pathologist to perform field selection before image analysis can be performed. These advancements are making quantitative image analysis a valuable reporting tool for pathology departments.
At the University of Pittsburgh Medical Center, we were interested in providing a quantitative CD8 cell count for our oncologists. We were approached to develop quantitative image analysis for CD8 cells within oropharyngeal squamous cell carcinoma. We began with a tissue microarray (TMA) from multiple tumours. We performed a CD8 immunostain on the TMA tissue section. Each core was manually counted, and this was used to assess the reliability of quantitative image analysis for the cores. The image analysis algorithm was optimised, and the parameters of the image analysis algorithm were fixed. Next, this same algorithm was used to quantify CD8 concentrations within a TMA section that had been previously quantified using a different method (AQUA). The results provided a similar predictive ability between the two methods. Based on these findings, we then decided to use the same quantitative algorithm on whole slide CD8 immunostain sections from a population with known outcomes. Based on the outcomes of the population, we were able to establish cut-offs associated with good vs poor prognosis. The unique aspect of this analysis was that we performed it on a whole tissue section – no fields for selection were necessary.
This represents a major improvement in the workflow because it does not require a pathologist to intervene prior to performing the image analysis. This however does not remove the pathologist from the process. We generate a procedure for performing the analysis and then an image analysis procedure report is signed out by a pathologist. This sign-out pathologist has the obligation to review the analysis for accuracy and confirm the results. The number of cells quantitated in this method can reach 10-100 thousand cells. This amount of cells would be impossible to count by manual (analogue) method. We have elected to divide the number of cells by the area analysed in order to generate the density of the CD8 inflammatory infiltrate. Providing the area measurement is also a task that is difficult by manual (analogue) methods. Historically, a high-power field is often recommended but variability within microscope components has added more complexity to using this standard measurement area. It is very easy to generate measurements from digital slides. Therefore, the only method that we could use to generate our results was to use automated image analysis. By using the entire tissue section, we have also eliminated selection bias introduced by choosing fields to analyse.
We have continued working to expand our image analysis in other types of malignancies. Many papers have been published about the quantification of CD8 cells in colon carcinoma. We have begun work on using quantification of CD8 cells within colon carcinoma. The system that we use for image analysis is a commercially available system that provides the image analysis tools and so we feel that our work for quantifying CD8 cells represents a more generalised method than those published previously. Although it is unlikely that every pathology department will perform quantitative image analysis, the technology should only be available from a few commercial entities. We have also worked to expand quantitative image analysis of CD8 cells in other types of solid malignancies.
Quantifying CD8 cells was initially described as a positive prognostic factor and our initial study found similar findings. However, immune based therapies are now being used for solid malignancies and the quantification of CD8 cells within the solid malignancies may have relevance to using immune based therapies. We continue to explore the possible role that quantification of CD8 cells may have in immune based therapy – either in determining the type of therapy to use or response to therapy. Although CD8 has an expansive amount of literature about its relevance in solid malignancies, there may be other immune cells that have relevance in predicting immune based therapy. These immune cells may already be known, or they may yet to be discovered. Given the nature of the immune system, digital quantification of the cells will be critical to using these cells in diagnostic reporting.
Immuno-therapy is changing the practice of oncologic treatment. As more immunotherapies are released, it will become important to assist our clinical colleagues with deciding which therapy to use in order to optimise patient outcomes. We have developed an automated quantification method for CD8 cells on whole tissue sections that can aid in the prognostic assessment of solid malignancies. We are using this same method to further evaluate whether it can be used in prediction of the types of therapies that should be used for solid malignancies.