Macromolecules (proteins, nucleic acids, lipids and carbohydrates) can be rapidly visualized in cells and tissue via staining with antibodies and/or special stains, followed by bright field color imaging. However, the quantitative analysis of such images is often hindered by variations in sample preparations, the limited dynamic range of color cameras, and the fact that image formation is not at a specific excitation and emission frequency, which is the hallmark of fluorescence microscopy. Through consistent sample preparation, fixation and imaging, we suggest that the signals associated with a macromolecule can be decomposed in the color space, and can render a scoring value on a cell-by-cell basis. Following this protocol, protein, lipid and DNA complexes are visualized with antibodies and special stains, and then imaged with a color CCD camera attached to a microscope.The key contributions of this article are in: (i) formulating the color decomposition as a global optimization problem, (ii) representing the signal complexes, associated with protein localization, with multiple prior models and (iii) applying the proposed method to the analysis of an end point on a cell-by-cell basis. In this context, global optimization is realized through the graph cut method, multiple prior models are specified through user initialization, and signal analysis, on a cell-by-cell basis, is established through a best effort in establishing cellular boundaries. The logical flow of these various computational steps is shown in , whereby the user first specifies regions associated with positive staining in an image, the nuclear regions are then automatically detected as a dark elliptic region (Yang and Parvin, 2003
), and are later further refined following color decomposition. The morphology and position of nuclear features allow the region-based tessellation of the image, and the subsequent scoring of the signaling complex on a cell-by-cell basis.
Computational steps in quantifying stained samples: in a single image, the user initializes the stained region associated with a signaling macromolecule. Learned parameters are subsequently used for the rest of the dataset.
We applied our method to fibroblasts grown from histologically normal breast tissue biopsies obtained from women from two distinct populations. The biopsies were digested in solution and the fibroblasts purified and grown in vitro. These fibroblasts were then grown under conditions that support adipocyte differentiation for 5–7 days before being fixed and stained with hematoxylin and Oil Red O, which stain DNA and lipids, respectively. Although hematoxylin and Oil Red O visualize nuclei in blue and lipids in red, respectively, there is still some overlap in the color space.
This article has been organized as follows: Section 2
reviews previous research in the area of color decomposition from histologically stained tissues; Section 4
demonstrates the effectiveness of our method when stains are co-localized; Section 3
provides the details of our method; Section 5
summarizes the results of our method and the application of our method to a large dataset and Section 6
concludes the article.