Evaluation of an automatic image classifier for analysis of bacterial growth on a multiple-agar plate developed for bovine mastitis.

Rapid identification of mastitis-causing bacteria is crucial for effective treatment decisions. Several multi-media agar plates have been developed to aid pathogen identification on farms or by veterinarians, but these methods require trained operators. Advances in AI-based automatic image analysis have shown potential for detecting bacterial growth on agar plates in both agriculture and medicine. This study aimed to evaluate the accuracy of an AI-based image bacterial classifier compared to a gold-standard laboratory assessment. A secondary objective was to examine how sample transportation affects diagnoses by comparing results from an on-farm bacterial classifier with those from a laboratory-placed classifier. A total of 1,299 milk samples were collected and analysed at the Swedish Veterinary Agency's Mastitis Laboratory using both accredited laboratory standards and the bacterial classifier. The image classifier is capable of identifying growth of eight different bacteria types on SELMA + multi-agar plates. Out of 1,212 samples that met the analysis criteria, the bacterial classifier provided diagnoses for 70%, while 30% required further evaluation. The classifier demonstrated high specificity for all diagnoses and high sensitivity for common pathogens such as Escherichia coli, Staphylococcus aureus, and non-beta-haemolytic streptococci, though sensitivity was lower for less common pathogens. In a subset of samples analysed by both on-farm and in-lab classifiers and the Mastitis Laboratory, 62% showed consistent diagnoses. The average transportation time was 4.9 days, which influenced bacterial growth. Interestingly, fewer mixed infections were detected post-transport. Automated image classifiers, like Bacticam, hold promise for on-farm mastitis diagnosis, supporting targeted antibiotic treatment and reducing antimicrobial use.