Identification of rare cell phenotypes in Sarcoma H&E by a combination of deep learning and image processing approaches

Abstract

Rare cells, i.e., those with a lower abundance within a population, play an important part in the conditions of health and disease of an organism. In Sarcoma, the presence of these cells can reveal the state of immune response or angiogenesis among other conditions [1]. Thus, its identification is crucial and one way to identify rare cells is through its phenotype [2]. Pooling of rare cells from large whole slide images (WSI) for training is tedious and time-consuming (Fig. 1). Deep learning techniques [3, 4] have been widely used for cellular segmentation and identification. However, an important pre-requisite of deep learning mandates large number of manually annotated training data, which is not always available.