Cell Segmentation Using Binarization and Growing Neural Gas

In corneal endothelium tissue engineering, automatic judgment is vital to determine whether cultured cells are suitable for transplantation, which can be achieved by measuring indicators from cell images. Indicator measurement requires an accurate image processing method for cell segmentation. We previously propose the system that combines simple image-processing filters suitably by genetic programming. However, it is too difficult to obtain high accuracy because of over-fitting. Therefore, achieving segmentation by an unsupervised learning method is an essential requirement for applying segmentation to several types of images. In this paper, we propose an unsupervised learning segmentation method using binarization and growing neural gas. This method segments the cells by performing vector quantization of cell borders and connecting units. The proposed method is comparable to a previously reported method. The results show that the proposed method has superior accuracy.