Digital Histopathology of Cancer

Cancer is a significant and growing public health concern. According to the World Health Organisation's estimates it is – after cardiovascular diseases – the second leading cause of death worldwide. Excluding non-melanoma skin cancers the most common types of cancer are for women breast cancer and for men lung cancer followed by prostate cancer.

While the biological understanding of cancer has expanded, so too has the selection of available treatments. More than one fourth of all new medicines entering the market during 2010-2018 were for treating cancer. In order for a patient to benefit from the wide variety of cancer treatments, and avoid adverse effects, their unique cancer has to be matched with the appropriate treatment. For this the cancer needs to be both diagnosed accurately and classified in detail.

Although non-invasive imaging methods, such as magnetic resonance imaging, have evolved substantially in recent years, the basis of cancer diagnosis is still in histopathology, that is, the pathological evaluation of tissue removed through surgery or needle biopsy. The light microscope has remained the pathologist's main diagnostic tool for a century and a half allowing for the examination of tissue down to cellular – and even subcellular – level. Important adjuncts to routine histopathological staining of tissue, needed for light microscopy, are techniques allowing for the visualization of protein antigens and nucleic acid in the tissue. These techniques, among which are immunohistochemistry and in situ hybridization, respectively, can be used for instance in the molecular characterization of cancer.

There are challenges in meeting the need for accurate diagnosing and characterization of cancer. One such challenge is posed by the shortage of pathologists observed in Finland and elsewhere. Another challenge is the variability in the interpretation of the tumor growth pattern (grading, such as Gleason grading in prostate cancer) and in the interpretation of certain tissue staining patterns (such as the immunohistochemical staining of the HER2 molecule in breast cancer). This variability manifests itself both between pathologists (interobserver variation) and also in the same pathologist's work over time (intraobserver variation). A third challenge is presented by the fact that the light microscope – although a reliable, cheap, and easy-to-use diagnostic tool – has shortcomings in the modern day pathology service.
Digital histopathology presents a new way of carrying out the central task of a pathologist in managing cancer patients, namely making the diagnosis and characterising the tumor in detail. Making the shift from a light microscope to a computer environment offers many benefits, some of which have been examined in this dissertation.
The present study was carried out with the purpose of developing and testing applications of digital pathology in order to improve the histopathological diagnosis of cancer. The individual studies looked at advancing the teaching and standardization of Gleason grading or prostate cancer, aiding in the interpretation of immunohistochemical staining of prostate and breast cancer, as well as facilitating the implementation of digital pathology by way of a novel whole slide image optimised image compression algorithm and mapping the determinants of an optimal imaging resolution for a whole slide scanner.
We demonstrated that whole slide images can be used to assess the Gleason grade of a prostate biopsy and that the use of an internet based platform can be beneficial in assessing interobserver variation in the grading and teaching and standardising the grading.
Besides Gleason grading another important aspect of prostate histopathology is the interpretation of immunohistochemistry. We created a method of viewing two whole slide images simultaneously and synchronously and tested this method in visualising the AMACR-p63 double stain along with normal hematoxylin and eosin staining of prostate biopsies. We showed that this technique can be used for histopathology education as well as in clinical diagnostics in selected cases.

A key issue in breast cancer diagnostics is defining the HER2 status of a tumor, that is, whether the tumor overexpresses the molecule and can then be treated with HER2 antibody based drugs. We studied the use of digital image analysis, using both photomicrographs and whole slide images, in aiding the pathologist in defining the HER2 status on a breast cancer surgical resection specimen. We showed that using a free and publicly available image analysis software can help to resolve cases otherwise deemed equivocal by conventional light microscopy.

The introduction of digital histopathology into routine diagnostic work is underway. One technical challenge is managing the large amounts of image data generated by whole slide images. When there is a need to store large numbers of whole slide images it is essential to strike a balance between image fidelity and file size. To deal with this issue we studied the optimal imaging resolution of a whole slide scanner using a methodology that can be utilised for instance in comparing whole slide scanners before acquiring one. In addition we introduced a novel way of image compression suited for whole slide images in order to reduce the storage footprint, and cost, of whole slide images.

The first two studies in this dissertation represent the very beginnings of whole slide imaging in pathology, and the field has advanced since then, perhaps in small part due to the findings in these studies. Taken together, the findings in this dissertation can hopefully advance the use of digital pathology in cancer diagnostics and thereby improve the care of cancer patients.