Empirical comparison of cell segmentation algorithms using an annotated dataset

Three popular algorithms (Viterbi active contour, watershed and a thresholding-based method) for segmenting cell nuclei are compared empirically. This is achieved via a large (20,000) image dataset that has been annotated by three independent nonexpert observers. The Hausdorff distance is used as an error metric and Chalana's modified Williams' index is used to compare each algorithm's output to the observers' data.