Extraction of fibers and nuclei of hepatic histopathologic specimen stained with silver and HE

In histopathologic diagnosis of hepatocellular carcinoma, N/C ratio (nuclear area / cytoplasmic area) is one of useful information. Our final goal is to develop a support system that enables the user to easily estimate the N/C ratio. Because cell area (Cell) is the sum of nuclear area (N) and cytoplasmic area (C), N/C ratio is usually calculated for each cell as N/(Cell-N). However, automatic calculation of N/C ratio has not been realized yet because automatic cell segmentation is very difficult. We, therefore, employed a different approach. Microscopic image of a hepatic histopathologic specimen is mainly composed of cell area and sinusoidal area. So, by discriminating them, the whole cell area is obtained. Also, by extracting the whole nuclear area, the N/C ratio is calculated. Cell area and sinusoidal area can be easily separated by extracting fibers which usually exist between them. It is also necessary to extract the positions of nuclei to calculate nuclear area. So, our intermediate goal is to extract both the fibers and the positions of nuclei from the same image. Silver staining strongly stains the fibers. It, however, does not stain the nuclei well. We combined silver staining with HE staining which stains the nuclei well, and employed this new staining method "silver-HE staining" to extract both fibers and nuclei. On extracting the fibers, it was difficult to discriminate between the fibers and silver granules because both of them are dark in blight-field image. This problem was solved by combining a dark-field image where the fibers are brighter than the silver granules. As a result, about 91.9% of correct ratio was experimentally obtained for extracting the fibers. Dark-field image was also used for extracting the positions of nuclei. Correct ratio was about 83.4% for the nuclei on which the silver granules are not overlapped too much.