Novel Color Normalization Method for Hematoxylin & Eosin Stained Histopathology Images

Abstract

With the advent of computer-assisted diagnosis (CAD), the accuracy of cancer detection from histopathology images is significantly increased. However, color variation in the CAD system is inevitable due to the variability of stain concentration and manual tissue sectioning. The small variation in color may lead to the misclassification of cancer cells. Therefore, color normalization is a very much essential step prior to segmentation and classification in order to reduce the inter-variability of background color among a set of source images. In this paper, a novel color normalization method is proposed for Hematoxylin and Eosin stained histopathology images. Conventional Reinhard algorithm is modified in our proposed method by incorporating fuzzy logic. Moreover, mathematically, it is proved that our proposed method satisfies all three hypotheses of color normalization. Furthermore, several quality metrics are estimated locally for evaluating the performance of various color normalization methods. The experimental result reveals that our proposed method has outperformed all other benchmark methods. INDEX TERMS Computer assisted diagnosis, color normalization, H & E stained histopathology image, fuzzy logic, quality metric.

Assessing the Impact of Color Normalization in Convolutional Neural Network-Based Nuclei Segmentation Frameworks

“…To assess the quality of CN quantitatively, the metrics that were introduced by Roy et al (2019) are used. Roy et al stipulated that the global color of the target image should be similar to that of the processed image.…”

“…The transformation is then used to match the histogram of the individual red, green, and blue channels of the query image to the red, green, and blue (RGB) channels of the specified image. A recent work demonstrates that histogram specification can effectively transfer the color of a reference image to a query image and is validated quantitatively (Roy et al, 2019). However, due to multiple dyes and tissue structures that vary from image-to-image, histogram specification is known to introduce image artifacts such as incorrect stain mapping.…”

“…Subsequently, the maximal range which comprises the stain content is estimated from the stain vectors and transformed back to the OD space. Various works state that while this method preserves histological information, normalization is not robust when images contain large stain variations (Bejnordi et al, 2016; Roy et al, 2019). Furthermore, because this method only addresses stain variation, other causes of color disagreements may not be addressed (Li and Plataniotis, 2015).…”

“…However, NMI is more indicative of an image's intensity information rather than the color content. Therefore, in addition to NMI, we utilize other color metrics introduced in Roy et al (2019) as well as normalized median hue (NMH), which is a novel proposed color consistency metric that measures variability in the hue across datasets.…”

“…The absolute difference is taken for these metrics and averaged by the total number of windows, W . A low AMCE value indicates similar color content between the target and processed images (Roy et al, 2019). Roy et al stipulates that the global color of the reference image should be approximately equal to the global color of the processed image, which should be captured by the AMCE.…”