Segmentation and Classification of Mast Cells in Histological Images with Deep Learning

摘要

Mast cells are an important part of the human immune system. They are also found in pathological pathways of various diseases. Their activity in tissues may be analyzed with a histopathological examination. In this preliminary study, we apply a deep learning approach for analysis of an original dataset of mast cells stained by toluidine blue. UNet and a convolutional neural network are applied for cell segmentation and classification, respectively. The result allows for the proposal of a methodological consideration for the subsequent full study. Basic deep learning approaches reach 66.64% segmentation accuracy when measured with the Dice coefficient and 81.36% class accuracy when the training dataset is measured with the categorical accuracy metric.