Performance Analysis of Color Normalization Methods in Histopathology Images

Abstract

Color normalization in histopathology is a prominent research topic in the image processing field as color in histopathology images plays a crucial role in diagnosis. As computer-aided diagnosis emerged, color normalization is much crucial as it becomes the foundation of medical image processing to assure algorithm precision and accuracy. In this paper, the main objective is to perform an analysis on three commonly used color normalization methods, namely histogram matching, histogram equalization, and stain unmixing methods. 60 breast histopathology images were used for testing purposes. Four statistical metrics were calculated to measure and determine the applicability of the color normalization methods: Structural similarity index measure (SSIM), Pearson’s correlation coefficient (PCC), visual saliency-induced index (VSI), and Gradient similarity (GS). Based on the outputs, it is found that the stain unmixing method demonstrates better than that of the histogram matching and histogram equalization methods with higher values in SSIM and VSI, and comparable values in PCC and GS.