We compare manually segmented masks of cell images that are independently hand-selected by several different people, with a new semi-automated method for estimating reference data for images of cells with very sharp clear edges. By quantifying the large differences in the hand selected masks, we validate the need for better precision. Mimicking the human eye, we select cell edges by a similar intensity differential used by the eye, but eliminating the human error that appears to arise. Resulting masks in our test of 16 images containing 71 cells, show consistently better results with the semi-automated method, and produce better results than using a kmeans-5 segmentation to estimate reference data.
Proceedings Title: 18th International Conference on Intelligent Systems for Molecular Biology
Conference Dates: July 11-13, 2010
Conference Location: Boston, MA
Pub Type: Conferences
reference data, segmentation