Melanoma is one of the most aggressive forms of skin cancer, causing a large proportion of skin cancer deaths. However, melanoma diagnoses by pathologists shows low interrater reliability. As melanoma is a cancer of the melanocyte, there is a clear need to develop a melanocytic cell segmentation tool that is agnostic to pathologist variability and automates pixel-level annotation. Gigapixel-level pathologist labeling, however, is impractical. Herein, we propose a means to train deep neural networks for melanocytic cell segmentation from hematoxylin and eosin (H&E) stained sections and paired immunohistochemistry (IHC) of adjacent tissue sections, achieving a mean IOU of 0.64 despite imperfect ground-truth labels.

One of the key areas in which artificial intelligence and the information sciences can contribute to biology is by helping human scientists understand cellular behavior in the context of a complex organism1,2. The utility of these methods is their ability to find novel regulatory interactions10 and even novel necessary regulatory genes11. These methods are indeed becoming indispensable for understanding the complex coordination of signals necessary to develop and maintain correct body shapes and organs. Moreover, such methods are required in order to develop interventions to make rational changes to complex anatomy and physiology, in the context of regenerative medicine and systems-level diseases such as cancer12. The coordination of cellular behavior towards the anatomical needs of the host organism, and away from tumorigenesis, is achieved in part via bioelectrical communication among many cell types13,14,15,16,17,18,19. Recent work showed that depolarization of resting potential in a special cell population in Xenopus embryos, so-called instructor cells, results in a metastatic-like conversion of normal melanocytes20.