Effect of Color Normalization on Nuclei Segmentation Problem in H&E Stained Histopathology Images

Cancer is one of the most common causes of death
today. Early diagnosis is of great importance for treatment.
Early diagnosis is made by detecting and examining cancerous
nuclei. For this reason, studies that automatically perform nuclei
segmentation on histopathology images are given wide coverage
in the literature to assist pathologists. There may be differences
in histopathology images due to staining or external factors.
In order to eliminate these differences, it becomes necessary
to perform color normalization on the image before starting
the segmentation process, both to standardize the images and
to improve the results obtained. In this study, the effects of
three different color normalization processes on the image were
compared. The comparison process was carried out according to
the nuclei segmentation results obtained by applying two different
deep learning models, U-Net and Residual U-Net. As a result of
the study, it has been observed that the training with the images
with the color normalization process gives more successful results
than the training with the original image