Automated AI-assisted colony counting may improve standardization in digital microbiology, but performance can be affected by colony density, culture medium, colony morphology, adhesion, and plate artifacts. We evaluated the Starry-300 AI colony counting system using 382 standardized pure culture bacterial and yeast plates across four agar media. AI-assisted counts were compared with a three-reader median ImageJ-assisted manual comparator derived from independent counts by experienced technologists. The AI workflow showed close agreement with the manual consensus comparator across a broad colony density range. Overall, 360/382 plates (94.24%) were within ±10 CFU and 377/382 plates (98.69%) were within ±30 CFU of the manual median count. Error-based and agreement analyses showed a mean absolute error of 3.19 CFU/plate; both the intraclass correlation coefficient and Lin’s concordance correlation coefficient were0.99. AI software analysis required approximately 5–15 s/plate, although this did not include plate handling, correction, or reporting. These findings support the analytical feasibility of reviewable AI-assisted colony enumeration under controlled pure culture conditions. Further validation using primary clinical specimens, mixed cultures, near-threshold samples, and external sites is required before broad clinical implementation.
Li et al. (Tue,) studied this question.