Comparison of Stain Normalization Methods for Acute Myeloid Leukemia Blood Slide Image

Color normalization is a crucial pre-processing method used in histopathological image analysis. The objective of this method is to reduce the color variations caused by staining procedures, which can affect the performance of automated image analysis algorithms. These color variations may occur from the inconsistency in protocols across laboratories during slide preparation. Besides, differences in the type and calibration of microscopes and cameras may cause discrepancies in image color. Inconsistent image acquisition settings, such as exposure time, white balance, and contrast adjustments, can also lead to variations in color representation. This paper provides a comparative analysis of three conventional color normalization methods: Reinhard, Macenko, and Vahadane. Reinhard approach uses mean and standard deviation normalization in the LAB color space to create a target image with the same color distribution as the reference image. Macenko uses SVD to separate stain vectors and then projects the picture onto a reference distribution to further normalize it. Vahadane approach combines non-negative matrix factorization with sparse stain separation for more robust and accurate stain normalization. The current study evaluates these methods based on five quantitative measurements which comprises of structural similarity index (SSIM), mean squared error (MSE), image entropy, correlation coefficient, and peak signal-to-noise ratio (PSNR). Our findings indicate that although Reinhard is the oldest color normalization method, it still provides the best result compared to other methods to normalize leukemia blood slide image.