ESCMID 2024: Artificial Intelligence Used to Reduce Diagnostic Microbiology Backlog

The Black Country Pathology Services is a network of laboratories in the Midlands, UK, covering a vast region historically known for its industrial background. In 2013, the lab moved into a custom-built facility designed for its operations, featuring advanced technology and automation, like the WASP lab, in its open-plan bacteriology section. It also has space for manual post-analytical processes. The pathology services established in 2018 began consolidating microbiology staff into a single site in 2020 during the COVID-19 pandemic, which was completed by September 2022, centralizing staff from five sites. The lab now operates 24/7, with night shifts focusing on urgent work and blood cultures. The consolidation also led to the adoption of Phenomatrix, an AI-based system for processing high-volume samples like MRSA and urine screens. As a result, the central lab in Wolverhampton, with 126 full-time equivalent staff, saw its capacity grow to handle 4,000 samples daily, a three-and-a-half-fold increase. Despite the workload surge, staff numbers did not decrease, but the approach to work shifted to manage the increased demands more efficiently.

Phenomatrix is AI-based software that analyzes the morphology, color, and quantification of growing colonies to interpret laboratory images. It employs specific algorithms to categorize these images, which users can customize. The software includes a human review step, allowing personnel to assess AI-generated results, make adjustments, and confirm the final output. Once approved, results are sent to the Laboratory Information Management System (LIMS), where negative outcomes are released automatically, while positive ones are flagged for further investigation. The development of Phenomatrix involved creating custom algorithms from local protocols, beginning with process mapping and using a large image collection to train the AI. This training helped the software accurately distinguish between positive and negative results and quantify them. The algorithms were refined for optimal performance, with an on-site verification ensuring accuracy and reliability.

The Phenomatrix uses specific cultures to achieve its results. For MRSA screens, the Phenomatrix rely on a blue colouration to identify positive results, with positive MRSA colonies showing a distinctive denim blue colour. Positive results are flagged for further testing. For urine screens, Phenomatrix uses chromogenic agar, specifically biome CPSE agar, in which it can identify mixed growth. The AI software considers the number and location of different colony types to determine the level of mixed growth. These cultures are crucial in processing large numbers of samples quickly. By automating the analysis of MRSA and urine screens, Phenomatrix helps clear sample backlogs, streamlines workflows and improves efficiency.

The study aimed to determine if Phenomatrix performed as expected and met the laboratory requirements. Data from a month of operation, collected a year after implementation, was reviewed to ensure complete integration. The outcomes produced by Phenomatrix were compared with the interpretations by laboratory scientists to assess accuracy. For MRSA screens in October 2023, Phenomatrix detected no false negatives. Some positive cases identified by the software were re-evaluated by scientists, revealing them to be negative, but the overall agreement between the software and human interpretation was 95%, with a sensitivity of 100%. For urine cultures, a similar review process was applied to a month of data, resulting in 100% agreement for negative results. However, the overall agreement for positive results was 92.7%, with some discrepancies caused by small or unclear colonies. Despite these discrepancies, the software's accuracy helped reduce manual examination time, allowing staff to focus on more complex cases.

Following a consolidation of services, night shifts were introduced to manage additional work depending on urgent cases and positive blood cultures. Simpler cultures were assigned for night reading due to reduced complexity and worker fatigue. The Phenomatrix system, which can process 100 plates per minute, significantly cut the time to read urine and MRSA screens, reducing reading time by over 90% and allowing more focus on complex tasks. The new software and workflows were smoothly implemented, with on-site support from team members. The software team accepted it due to its ease of use and immediate impact. The backlog of hundreds of unread samples was solved, which provided immediate benefits to the diagnostic lab. The plans are in place to expand the system to other screens, including Carbapenamase-Producing Enterobacteriaceae (CPE) screen, Methicillin-Susceptible Staphylococcus aureus (MSSA) screen, and Vancomycin-resistant Enterococci (VRE) screen. This expansion could bring further efficiency as more screens are integrated into the workflow.

European Society of Clinical Microbiology and Infectious Diseases (ESCMID) 2024, 27^th April–30^th April 2024, Barcelona, Spain