1. Introduction
Neuroendocrine tumor (NET) is one of the most frequent cancers worldwide. Computer-aided image analysis, like Ki67 counting, of digitized specimens could potentially support improved characterization of NET. Efficient and accurate cell segmentation is a prerequisite for many computer-aided quantitative analyses such as morphological feature extraction or cell recognition, which is critical for Ki-67 counting. Many state-of-the-art approaches such as multiple level set [1], supervised learning [2], and semi-supervised classification [3] have been successfully applied to nuclei/cell segmentation on microscopic images.