Modified Reinhard Algorithm for Color Normalization of Colorectal Cancer Histopathology Images
Santanu Roy, Shubhajit Panda, Mahesh Jangid · 2021 29th European Signal Processing Conference (EUSIPCO) · 2021
In recent trends, Computer Assisted Diagnosis (CAD) enables the pathologists to diagnose cancer disease from histopathology images very efficiently. Color normalization is a pre-processing step prior to cancer classification task which can reduce the computational complexity of the classifier. However, existing color normalization methods are fraught with the problems of data loss and huge computational complexity. The purpose of employing this color normalization method is to reduce the color variation among a set of histopathology images so that in the next step, the classifier can efficiently extract the prominent features for cancer grading. This color variation is generally occurred due to using different scanners, stain concentration variability and poor tissue sectioning, while preparing the histopathology slides. In this paper, a modified Reinhard algorithm is proposed for color normalization of Hematoxylin and Eosin (H&E) stained colorectal cancer histopathology images. The limitations of Reinhard algorithm are alleviated by the proposed algorithm. Moreover, a statistical analysis is provided to prove that proposed algorithm does not cause any data loss and subsequently, it satisfies all four hypotheses of color normalization. Furthermore, the performance of the proposed algorithm is compared with other existing color normalization methods both qualitatively and quantitatively.