Evaluation of an automated image analysis system for counting human tumor colonies

The Omnicon FAS II image analysis system was applied to counting tumor colonies grown in a soft agar human tumor clonogenic assay with a detailed protocol designed to assess the instrument's sensitivity, specificity, precision, and accuracy. Comparisons of technician and instrument counts were done on a blinded basis. Sensitivity studies (which used metal microspheres) yielded a correlation coefficient (r) of 0.999 between technicians and the counter. A field-by-field analysis of the instrument's specificity for identifying individual objects correctly as tumor colonies rather than artifacts (as identified by the technician) was excellent (r = 0.95). In the precision studies (determined with repeated automated counting of the same samples for five days), the median coefficient of variation was less than 7%. Accuracy was evaluated on cultures of fresh biopsies from 30 human cancers obtained for drug sensitivity testing as well as on a series of tumor cell lines. The correlation between the mean number of colonies counted by the technicians and by the colony counter was greater than 0.91. Similar comparisons of mean percent survival of tumor colony-forming cells after drug exposure between technician and machine were also quite acceptable (r = 0.85). We conclude that the colony counter provided sufficient reliability to be applied to counting human tumor colonies grown in vitro. In addition, the colony counter performed the Petri dish counts ten times faster than experienced technicians and without associated operator fatigue.