Automatic cell analysis for P53 immunohistochemistry in bladder inverted papilloma

The automatic cell analysis system developed in the present study has two merits. Firstly, a nonlinear hybrid color space is built for pixel classification. The components of the nonlinear hybrid color space are selected from RGB, HSI, XYZ, L*u*v*, and nonlinear polynomial combinations of these components, with the objective of achieving large class separability between nuclei and non-nuclei pixels. Secondly, a novel segmentation algorithm for overlapping or touching nuclei is developed by incorporating hierarchical clustering into distance transform to locate nuclei centers. The effectiveness of the proposed methods is tested using a few images.