Work

AI-driven identification and analysis of inhibition zones in disk diffusion tests with the hue contrast method

Year: 2024

Type: article

Abstract: Antibiotic susceptibility testing (AST) is crucial for monitoring bacterial resistance and guiding treatment. The Kirby–Bauer disk diffusion test, a standard AST method, encounters challenges due to variability in manual measurements. We developed an automated imaging system with a reporting function for disk diffusion tests to address these issues. A disk diffusion test was carried out on Salmonella enterica with ceftiofur (CF), enrofloxacin (ER), and colistin (CT) disks. The disk images captured by a smartphone were annotated and processed on the Roboflow platform and then used to train the YOLOv7 model for text recognition. The YOLOv7 model achieved 100 % recognition accuracy for the CF and CT disks and 98 % for the ER disks. We employed the Hough circle and hue contrast methods to detect Petri dish and inhibition zone contours. The hue contrast method achieved an 88 % success rate for detecting inhibition zones, outperforming the Hough circle method, which showed a success rate ranging from 54–82 %. We compared the results obtained from the imaging system with those of manual measurements. The mean differences in the inhibition zone diameters were −1.3 mm for CF, −1.0 mm for ER, and −0.5 mm for CT for Salmonella enterica. The accuracy and consistency of the imaging system were also shown in Escherichia coli with mean differences of 0.1 mm for CF, −1.5 mm for ER, and 0.15 mm for CT, with all the data points within the 95 % limit of agreement. In conclusion, our automated system, which uploads images to a cloud server for analysis and generates AST reports via the LINE chatbot, significantly improves interpretation accuracy and reproducibility.

Source: Microchemical Journal

DOI: 10.1016/j.microc.2024.112459