Automatic hepatocellular image analysis for morphological parameters

Liver cancer is one of the ten major causes of death for male and female in Taiwan. Therefore, early detection of hepatocellular carcinoma is very important for early treatment. In clinics, the routine confirmation of hepatoma is always based upon the biopsy analysis. It is heavily dependent on physician's experience and training. Hence, the purpose of this study is to develop an objective and quantitative liver tissue analysis system. In this study, edge based and threshold based methods are both employed to segment the nuclei for the liver tissue image. The morphological features are then extracted using Hough transform. With these parameters, the multilayer perceptron is then applied to find the contours of nuclei. The accuracy for nucleus extraction is about 91%. Finally, the morphological parameters of nuclei needed in the biopsy analysis, such as the averaged diameter and their variations, may be evaluated automatically.