Automated counting of mammalian cell colonies

Investigating the effect of low-dose radiation exposure on cells using assays of colony-forming ability requires large cell samples to maintain statistical accuracy. Manually counting the resulting colonies is a laborious task in which consistent objectivity is hard to achieve. This is true especially with some mammalian cell lines which form poorly defined or 'fuzzy' colonies, typified by glioma or fibroblast cell lines. A computer-vision-based automated colony counter is presented in this paper. It utilizes novel imaging and image-processing methods involving a modified form of the Hough transform. The automated counter is able to identify less-discrete cell colonies typical of these cell lines. The results of automated colony counting are compared with those from four manual (human) colony counts for the cell lines HT29, A172, U118 and IN1265. The results from the automated counts fall well within the distribution of the manual counts for all four cell lines with respect to surviving fraction (SF) versus dose curves, SF values at 2 Gy (SF2) and total area under the SF curve (Dbar). From the variation in the counts, it is shown that the automated counts are generally more consistent than the manual counts.