A Sparse Feature Clustering-based Stain Normalization Method for Pathological Images
Qian Bian, Elcid A. Serrano · 2023
Gold standard for diagnosing malignant tumors is pathological diagnosis, and digital pathology is current trend. Difference in stain process of pathological slides and lack of standardization in digitization of pathological images have posed challenges to development of digital pathology due to stain variation in images. Therefore, this paper proposed a new stain normalization method for pathological images based on sparse feature clustering. Based on CycleGAN, this method effectively addressed the problem of inadequate representation of template image in traditional stain normalization methods. Simultaneously, by adopting a sparse K-means clustering algorithm, this method effectively handled significant stain variations in datasets. In comparison, SSIM and PSNR of this method were 0.96 and 32.62, respectively. These values were significantly superior to those of Reinhard and Vahadane methods. Moreover, this method obtained better visual effects, which would provide guidance for auxiliary clinicopathological diagnosis.