Self-organizing map-based multi-thresholding on neural stem cells images

Automatic segmentation and tracking systems can be useful tools for biologists to monitor and understand the proliferation and the differentiation of neural stem cells. This paper applied the self-organizing map-based multi-thresholding on the neural stem cells images. Using local variance as the local spatial feature and quadtree decomposition as the sub-sampling method, inner-cell regions, cell borders and background can be roughly classified. Based on these results, proper foreground and background seeds were constructed for the seeded watershed segmentation and every single cell in a cell cluster can be segmented correctly. The results were also compared to the seeded watershed segmentation based on regional maxima method.