Experimental Assessment of Color Deconvolution and Color Normalization for Automated Classification of Histology Images Stained with Hematoxylin and Eosin

Abstract

Histological evaluation plays a major role in cancer diagnosis and treatment. The appearance of H&E-stained images can vary significantly as a consequence of differences in several factors, such as reagents, staining conditions, preparation procedure and image acquisition system. Such potential sources of noise can all have negative effects on computer-assisted classification. To minimize such artefacts and their potentially negative effects several color pre-processing methods have been proposed in the literature—for instance, color augmentation, color constancy, color deconvolution and color transfer. Still, little work has been done to investigate the efficacy of these methods on a quantitative basis. In this paper, we evaluated the effects of color constancy, deconvolution and transfer on automated classification of H&E-stained images representing different types of cancers—specifically breast, prostate, colorectal cancer and malignant lymphoma. Our results indicate that in most cases color pre-processing does not improve the classification accuracy, especially when coupled with color-based image descriptors. Some pre-processing methods, however, can be beneficial when used with some texture-based methods like Gabor filters and Local Binary Patterns.

Spatial immunoprofiling of the intratumoral and peritumoral tissue of renal cell carcinoma patients

“…We estimated that our lymphocyte classifier would perform better in cell detection and classification after color normalization possibly as it was developed with a random trees algorithm. However, convolutional neural network-based image analysis could improve classification due to lower sensitivity for technical color variations [ 41 ]. Moreover, we found immune excluded tumors to associate with fewer metastatic organs and superior overall survival in our primary dataset, representing a potential clinical biomarker in RCC patients.…”

Deep Learning Prediction of Metastasis in Locally Advanced Colon Cancer Using Binary Histologic Tumor Images

“…The preparation of the black and white images could be a source of interobserver variability, but we assume that they might be better suited for reproducible deep learning analyses between different centers by avoiding the considerable differences in H&E staining results of different laboratories [ 42 ]. Up to now, no standardized staining platform for H&E has received wide acceptance, and in most cases color pre-processing does not improve classification accuracy [ 43 ].…”

The utility of color normalization for AI‐based diagnosis of hematoxylin and eosin‐stained pathology images

“…Others investigated only one or two methods or a limited number of images and datasets, and therefore lack generalizable conclusions [18,23,27,29,33,35,36]. Recently, Bianconi et al [37] compared multiple methods on various cancer types but only reported the classification efficacy on microscopic regions of sections (e.g. 256 Â 256 pixels; referred to as patches) sampled from WSIs, as opposed to slide-level results which would be translatable and applicable to patient diagnosis.…”

“…256 Â 256 pixels; referred to as patches) sampled from WSIs, as opposed to slide-level results which would be translatable and applicable to patient diagnosis. Furthermore, the effect of color normalization for classifier performance is primarily studied in the context of images in which both training and testing sets originate from one center, such as a single hospital or research institute [23,26,34,35,37]. Single-center color inconsistencies between WSIs are mainly due to sample processing and fading over time.…”