Quality Assessment of Color Normalization Method by Similarity Index Metrics-A Comparative Study for Histopathology Images

Abstract

One of the biggest challenges of histopathology image processing is to preserve structural similarity while processing for further research. Color normalization algorithms can play a significant role in preserving the structure of histopathology images from various standpoints. In this research, we provide a comparative analysis of seven distinct color normalization algorithms by evaluating three state-of-the-art structural similarity index metrics often employed in image processing. 100 malignant prostate cancer histopathology tissue images (256 x 256) from various grading (Gleason score 3, 4, and 5) have been utilized here. The structure similarity index matrix (SSIM), quaternion structure similarity index matrix (QSSIM), and multi-scale structure similarity index matrix (MS-SSIM) are three state-of-theart quality evaluation metrics used in this research. Also, by computing the mean standard deviation (SD) of the grayscale images to determine the noise level and signal-to-noise ratio (SNR), respectively, we examined six denoising algorithms with various parameters to improve the efficacy of this analysis. This study provides a higher value for each of the three-similarity metrics, indicating a relatively superior performance for the Blind Color Decomposition algorithm. Furthermore, the Gaussian algorithm outperforms the six denoising techniques in terms of SNR and SD. When we integrated the Blind Color Decomposition and Gaussian algorithm with our experimented specific parameters, we were able to obtain the ultimate higher value for all three structural similarity index metrics. We anticipate that this analysis will have a substantial impact on various aspects of histopathology image processing, including segmentation, classification, feature extraction, and the creation of novel algorithms.