Upfold, J. K.; van de Schoor, A.; Elvebakken, H. F.; Petersen, O.; Elvebakken, C. F.; Kustner, C.; Madsen, M.

Manual colony counting remains the rate-limiting, operator-dependent step in culture-based food microbiology quality control (QC). Automated colony analysis using machine learning (ML) offers the potential to standardise, accelerate, and improve the traceability of this process. However, systematic multi-method validation data for AI-based platforms against recognised international standards remain scarce. We conducted a prospective, multi-study validation of the Reshape Smart Incubator which is an automated imaging and ML-based colony analysis system, across eight ISO microbiological reference methods. In total, 887 plates were analysed, spanning qualitative (presence/absence) detection of Listeria spp. (ISO 11290-1) and Salmonella spp. (ISO 6579), and quantitative enumeration of total viable count (ISO 4833), Bacillus cereus (ISO 7932), Enterobacteriaceae (ISO 21528), coagulase-positive Staphylococci (ISO 6888), yeasts and moulds (ISO 21527), and lactic acid bacteria (ISO 15214). Automated results were benchmarked against the consensus of three or more trained technicians. The platform achieved 100% agreement with manual assessment for all both qualitative detection methods (ISO 11290-1, ISO 6579) with zero false positives and zero false negatives. For quantitative enumeration, agreement ranged from 92.97% (ISO 15214, n=122, using ISO-aligned {+/-}10%/>30 CFU thresholds) to 98.46% (ISO 21528, n=130). Where discrepancies occurred, they largely coincided with plates showing high inter-technician variability. Precision testing demonstrated a coefficient of variation of 5.88% and a mean standard deviation of 0.44 CFU for low-count plates. This study presents a comprehensive multi-ISO validation of an AI-based colony analysis system to date. The AI models demonstrated performance comparable to or exceeding that of trained human technicians across a broad range of microbiological targets, agar types, and colony morphologies, thereby supporting their use as a validated and traceable alternative to manual plate reading in accredited food microbiology quality control laboratories.