Data-Driven Cancer Research with Digital Microscopy and Pathomics

The study of cancer biology is a highly data-driven process with the intent to develop better diagnostic and treatment strategies. Broadly speaking, many novel avenues of cancer research that incorporate data from multiple modalities, such as demographics, clinicopathologic, genomics, laboratory testing, imaging, pharmacologic, and survival data, can be viewed within the paradigm of dynamic data-driven application systems (DDDAS). These various kinds of biomedical data are analyzed, integrated, comparatively evaluated, and constantly mined to help develop models for biomarker discovery that can be used to predict disease progression and treatment response. Findings from sophisticated data analyses then trigger further scientific inquiry, novel experiments, and observational studies to collect more data and explore the incorporation of data from additional modalities, which are used to refine previously generated models. Digital microscopy and Pathomics, which is referred to collectively as quantitative digital histopathology, play an increasingly important role in this manner in modern cancer research. Digital microscopy makes it possible to capture extremely detailed images of tissue microanatomy, cellular architecture, and subcellular structures from tissue samples on glass slides. Pathomics is the process of extracting large quantitative imaging feature sets from high resolution tissue images. Examples of Pathomics data include shape, color intensity, and texture properties of cancer nuclei, patterns of tumor-infiltrating lymphocytes, and characterizations of tumor regions. Quantitative digital histopathology is a DDDAS-based process in itself. In typical DDDAS settings, there exists a dynamic, data-driven interaction between simulations and experiments, where experimental data are used to refine simulation models and simulations help drive new experiments. In cancer research with quantitative data from tissue images, a similar interaction exists between imaging data (digital microscopy) and machine learning models (Pathomics analyses). Training data generated from tissue images are employed to train and refine image segmentation and classification models. Imaging features extracted by machine learning models.