(A-17) A Novel Color Deconvolution Method with open source software for Accurate Visualization of Tissue Component

Advances in imaging instrumentation and data management provide the foundation for computational approaches to analyze digitized color images of histological tissue sections. Color deconvolution, separating the spectrum of colors representing stained molecular or tissue compartments, is a crucial step in developing objective and quantitative analysis of tissue sections. The linear color deconvolution method, which decomposes color images into absorbance values of individual stains, has been widely used in the field of digital pathology. However, this method heavily relies on identifying optimal color vectors of corresponding stains, which is often challenging to identify for individual tissue images. Moreover, under polychromatic conditions, linear color deconvolution approaches demonstrate limited performance and struggle with more than three stained colors in tissue sections, such as multiplex IHC. To address these limitations, we have developed an effective and intuitive color deconvolution method that robustly and accurately separates more than three stain signals.

We established a novel deconvolution approach based on the 2D color 

spectrum map of the color images. Spherical coordinates were used to represent the absorption spectrum of color images. The radius coordinates represent the overall magnitude of the absorption and the theta and phi coordinates are associated with the color pattern of absorption. The spectrum map of the color images can then be represented using  a theta-phi joint histogram, which is created by calculating the frequency of pixels at different theta and phi intensities in the image. In the spectrum map of the color images, different stains will appear in distinct locations of the map and thus the stains can then be spatially segmented in the color map in order to visualize them.  The segmentation of the 2D color spectrum map can be performed either  manually or automatically through machine learning clustering algorithms.To facilitate the implementation of the proposed method, we have further packaged the proposed method as an imageJ plug-in.

We showed our method can effectively visualize the distinctive color spectrum in typical two stained components of H&E images or IHC images. The transformation of the segments of distinct stain/tissue components in the color spectrum map provides an effective and intuitive approach to deconvolute the colors. We also show that the proposed method can accurately decouple immunohistochemistry (IHC) staining with five chromogens to individual stains. We further packaged our method in ImageJ plug-in and demonstrated it provides an effective, and accessible interface for users to implement and customize the color deconvolution analysis on color images.

The proposed method offers a robust and effective solution to address color variations in histopathology images, improving the reliability and accuracy of computational pathology. By accurately visualizing individual tissue components, regardless of color differences, our method enhances the precision and usability of computational analysis in histopathology.

The integration of our method as a plugin for ImageJ enhances accessibility and usability, allowing researchers and pathologists to leverage its capabilities regardless of programming expertise. The user-friendly GUI simplifies the application of our method, promoting wider adoption within the computational pathology community.