Estimating the Number of Manually Segmented Cellular Objects Required to Evaluate the Performance of a Segmentation Algorithm

Adele P. Peskin, Joe Chalfoun, Karen Kafadar, John T. Elliott

Abstract: We propose a new strategy for estimating the number of cellular objects that should be manually segmented for evaluating the segmentation performance of an algorithm. The strategy uses geometric and edge quality measurements that are directly related to segmentation performance, but do not require highly accurate segmentation. Sample sizes are determined from standard deviations of cell features calculated from the entire image set. We examine the relationship between approximate confidence level and sample size. The use of our strategy may reduce the effort and time required for generating a reference dataset for evaluating segmentation algorithm performance with images of biological cells. We demonstrate the usefulness of this methodology on a large and diverse data set for which reference data are available.