Marked Point Process for Nuclei Detection in Breast Cancer Microscopic Images

The automatic detection of nuclei within cytological sample imagery is crucial for quantitative analysis in medical applications. Unfortunately, the classical segmentation algorithms perform poorly for cytological images if precise seeds of nuclei are not given in advance. To tackle this problem, we propose nuclei detection method based on Bayesian recognition framework. It finds spherical regions with intensity distribution characteristic for nuclei and approximates them by disks. The process of disk generation can be viewed as marked point process (MPP). To penalize disk overlap, we added priori distribution of configuration based on pairwise interaction. The best disk configuration maximizes the probability of configuration given image data and pairwise interactions between disks. Deterministic algorithm based on Steepest Ascent method was used to search the configuration space in order to find the solution. To test the effectiveness of the method, it was applied to recognize nuclei in cytological images of breast cancer tissue.