Drew Williamson, MD
Fellow in Clinical Informatics
Drew FK Williamson, Iain Carmichael, Richard J Chen, Faisal Mahmood
Research Category: Cancer
Intratumoral heterogeneity–variations between cells within the same neoplasm–is increasingly being recognized as a hallmark of cancer with implications for diagnosis, treatment response, and prognosis. Though it has primarily been elucidated through molecular methods such as DNA or RNA sequencing, heterogeneity of the histologic appearance of tumor cells has long been recognized as intrinsic to the clinical practice of pathology. Efforts to characterize the connections between the phenotypic heterogeneity that pathologists observe every day and the relatively well-studied ‘omic heterogeneity have been hampered by an unmet need for quantifying histologic heterogeneity for subsequent correlation. We have developed a deep learning-based system which utilizes features from individual cells and the tumor microenvironment to calculate a histologic heterogeneity score for a given slide. Using this score, we identify statistically significant correlations with genomic measures of intratumoral heterogeneity and patient outcomes when tested on patient samples from the Cancer Genome Atlas. With this score, intratumoral heterogeneity can be investigated with the cheap and readily available hematoxylin and eosin stained tissue that pathologists use to diagnose nearly every cancer patient, allowing for correlations to be discovered at an unprecedented scale.
Cancer cells within a tumor are often thought of as clones a single cell that have escaped the body’s normal controls on growing and dividing. Researchers have recently started studying the ways that these “clones” are different from each other and how that might affect diagnosis and treatment of cancer. Most ways to study these differences focus on the DNA of the tumor cells, finding that there are often distinct groups of cells that share some, but not all, of their DNA. When pathologists look at a patient’s tumor under the microscope, they often see something similar–that the cells in one area of the tumor can look subtly or very different from those in another area. We created a system to measure the visual differences between cancer cells under the microscope to catalogue these differences. We aim to help researchers with this tool start to figure out how to make new treatments that stop all cells in a patient’s tumor, not just the ones that look a certain way or that have a certain type of DNA.