Improving white blood cells image segmentation performance under domain shift

Abstract

Domain shift is a phenomenon in which the distribution of the input space is shifted due to a change in the description of the observed variable or in the observing system. In this paper, the structure of white blood cells is analyzed, U-Net semantic segmentation network is constructed, and image segmentation deep learning of white blood cells, especially their nuclei, is carried out on PBC data set. Using the learned weights, the segmentation results are tested on the Test-B data set of Raabin-WBC.