Digital Image Analysis in Pathology Using DNA Stain: Contributions in Cancer Diagnostics and Development of Prognostic and Theranostic Biomarkers

Abstract

Traditionally, histopathological evaluations of tissue sections are performed for the diagnosis and grading of cancers using subjective appraisal of the tissue and cell phenotypes. In the last decade, a combination of unprecedented advances in imaging and computing technologies and novel machine learning-based algorithms has driven the field of histopathology into a new dimension. Machine learning and deep learning methodologies applied on digitalized hematoxylin and eosin stained tissue sections have out-performed conventional methods to accurately identify cancer cells or classify tumors into prognostic groups. Nevertheless, we believe that a precise, standardized measurement of nuclear morphology and chromatin texture based on a DNA stain can further improve the diagnosis of cancers and identify patients with high-risk of recurrence, alone or in combination with other clinical, pathological or molecular information. Changes in the morphology of cell nuclei and tissue architecture are intrinsic characteristics and hallmarks of cancer. Nuclei from cancerous samples exhibit different morphological and chromatin texture than nuclei from normal cells, thus, reflecting the structural and molecular effects of genetic and epigenetic alterations driving cancer processes. By image analysis, the chromatin texture can be measured in high-resolution breaking down the components of the nuclear changes into multiple quantifiable units that can be studied independently and in combination using advanced machine learning methods. This allows the investigator to examine associations of such changes with cancer progression and clinical outcomes. There is now increasing interest in developing new algorithms and platforms to decipher spatial relationship between cell subpopulations in whole tissue sections.