Image dataset of disk diffusion assay scanned with the SIRscan system

We present a comprehensive dataset deposited in the DRYAD repository, which includes high-resolution images and corresponding automated interpretations from the SIRscan system. This dataset is intended to support the development and validation of machine learning models and other analytical tools aimed at enhancing the accuracy of antimicrobial resistance detection, particularly in Gram-negative bacteria. The images in this dataset were generated using disk diffusion methods following the EUCAST guidelines and encompass a variety of phenotypic resistance patterns against beta-lactam antibiotics.

The dataset includes 225 Gram-negative bacterial isolates with a total of 862 unique phenotypic categories, reflecting various resistance mechanisms, including extended-spectrum beta-lactamase (ESBL), plasmid-mediated AmpC beta-lactamase, and carbapenemase production. Each image is paired with an automated reading provided by the SIRscan system, which includes measurements of inhibition zone diameters and preliminary resistance classification. This pairing of raw image data with machine-generated interpretations offers a valuable resource for the development of advanced algorithms for antimicrobial resistance prediction and other related applications.

This dataset is part of an ongoing effort to provide open-access resources that can be used to benchmark and validate the performance of machine learning models in clinical microbiology. By sharing these data, we aim to facilitate the development of more accurate and efficient diagnostic tools, ultimately contributing to better clinical outcomes and more effective antimicrobial stewardship.

Methods
The dataset was collected as part of a larger study aimed at improving the detection of antimicrobial resistance in Gram-negative bacteria using disk diffusion methods. The data collection process involved the following steps:

Isolate Selection: A total of 225 Gram-negative bacterial isolates were selected for the study. These isolates were chosen based on their clinical relevance and diversity in resistance mechanisms, including extended-spectrum beta-lactamase (ESBL), plasmid-mediated AmpC beta-lactamase, and carbapenemase production. The isolates represented a range of species commonly encountered in clinical microbiology.

Disk Diffusion Testing: Disk diffusion tests were performed on each isolate following the standardized guidelines provided by the European Committee on Antimicrobial Susceptibility Testing (EUCAST). The tests involved placing antibiotic-impregnated disks on agar plates inoculated with the bacterial isolates. After incubation, the diameter of the inhibition zones around each disk was measured to assess the susceptibility of the bacteria to the antibiotics.

Image Acquisition: High-resolution images of the agar plates were captured using the SIRscan automated system. The SIRscan system is equipped with advanced imaging technology that allows for precise measurement of inhibition zones. Each image was stored with a unique identifier corresponding to the specific isolate and antibiotic combination tested.

Automated Interpretation: The SIRscan system also provided automated interpretations of the disk diffusion results. These interpretations included measurements of inhibition zone diameters and preliminary classifications of resistance mechanisms based on predefined algorithms aligned with EUCAST guidelines. The automated interpretations were recorded alongside the corresponding images.

Data Annotation and Validation: The collected images and automated interpretations were reviewed and validated by clinical microbiologists to ensure accuracy and consistency. Any discrepancies between the automated readings and expert evaluations were noted, and the dataset was curated to include only high-quality data.

Data Organization: The final dataset includes 862 unique phenotypic categories, each associated with a high-resolution image and its corresponding automated interpretation. The data is organized in a structured format, with metadata describing the isolate, antibiotic tested, inhibition zone measurements, and the interpreted resistance mechanism.

This dataset, now available in the DRYAD repository, represents a robust resource for researchers developing and validating machine learning models and other tools for antimicrobial resistance detection. The collection process was designed to ensure the reliability and relevance of the data for clinical and research applications.