Precise segmentation of nuclei in hepatic histological images

Nuclear morphometric features such as shape of nuclei are useful in diagnosing hepatocellular carcinoma, especially well-differentiated hepatocellular carcinoma (ewHCC). We previously developed a support system for diagnosing ewHCC that enables the user to estimate the nuclear density and the roundness factor of nuclei. In the system, the contours of a nucleus are extracted as a collection of discrete contour points which are converted to a spline curve by interpolating them. The user can correct wrong contours by moving the contour points using a mouse. So, the number of contour points is limited to a small number, usually eight, in order to reduce the time for contour correction. As a result, it was difficult to precisely express the shape of nuclear contours especially in deformed nuclei. In order to solve this problem, new process to improve the contours was introduced. After the contour correction using the GUI, the contours were improved using 80 contour points. An energy function including four energy terms, the energies of the image, the distance between contour points, the curvature, and the color differences, was used. Experimental results showed this process is effective for all the three types of contours, normal round-shaped, deformed, and vague contours. The average absolute error of contour positions was reduced to about 1/3 of the conventional error. As a result, the average absolute error of nuclear areas was reduced to about 1/10, which corresponds to 0.50% of the average nuclear area. The error of roundness factors was also improved. This new process is totally automatic, which means no additional manpower is required.