Edge Enhancement and Dual UNet Fusion Based GAN for Structure Preserving Stain Normalization

EasyChair Preprint 14662

Abstract

Histopathology is the diagnosis and study of tissue diseases, and staining is a crucial part of its analysis. However, differences in laboratory protocols and scanning devices can often result in significant variations in the appearance of images, imposing obstacles to the diagnosis process. To address this issue, we propose a method called EG-DUNet, which is a GAN-based dual UNet network combined with edge enhancement information. The EG-DUNet network is able to obtain multi-scale feature fusion, which helps capture the shape and structure of cells in tissue sample images. To optimize color consistency, a style loss constraint is incorporated into the proposed network. Compared with current mainstream methods, our experimental results show that the EG-DUNet achieves more competitive performance on the MITOSATYPIS-14 contest dataset.

Keyphrases: Computer Vision and Pattern Recognition, Histology Images, Medical Imaging, Staining Normalization, Style Loss, UNet, artificial intelligence and technology, deep learning, designed loss function

@booklet{EasyChair:14662,
  author    = {Yicheng Zhao and Jiacheng Lu and Bo Li and Hui Ding and Guoping Huo},
  title     = {Edge Enhancement and Dual UNet Fusion Based GAN for Structure Preserving Stain Normalization},
  howpublished = {EasyChair Preprint 14662},
  year      = {EasyChair, 2024}}