Computational normalization of H&E-stained histological images: Progress, challenges and future potential.

Different types of cancer can be diagnosed with the analysis of histological samples stained with hematoxylin-eosin (H&E). Through this stain, it is possible to identify the architecture of tissue components and analyze cellular morphological aspects that are essential for cancer diagnosis. However, preparation and digitization of histological samples can lead to color variations that influence the performance of segmentation and classification algorithms in histological image analysis systems. Among the determinant factors of these color variations are different staining time, concentration and pH of the solutions, and the use of different digitization systems. This has motivated the development of normalization algorithms of histological images for their color adjustments. These methods are designed to guarantee that biological samples are not altered and artifacts are not introduced in the images, thus compromising the lesions diagnosis. In this context, normalization techniques are proposed to minimize color variations in histological images, and they are topics covered by important studies in the literature. In this proposal, it is presented a detailed study of the state of art of computational normalization of H&E-stained histological images, highlighting the main contributions and limitations of correlated works. Besides, the evaluation of normalization methods published in the literature are depicted and possible directions for new methods are described.