Researchers at the University Medical Center Oldenburg have developed “CarbaDetector”, an AI model that identifies antibiotic-resistant bacteria more precisely than previous methods – with significantly fewer false-positive results. This reduces unnecessary, time-consuming and cost-intensive follow-up examinations. The results were published in “Nature Communications” on November 14, 2025.

Lead researcher Prof. Dr. Axel Hamprecht from the University Institute of Medical Microbiology and Virology at Oldenburg Hospital emphasizes the benefits: “Common screening methods often lead to false-positive results that require expensive additional tests.” Dr. Linea Katharina Muhsal, first author, adds: “CarbaDetector saves resources and accelerates diagnoses.”

The model targets carbapenemase-producing Enterobacterales (CPE), pathogens that destroy last-line antibiotics such as carbapenems and cause serious infections such as sepsis or pneumonia. Worldwide, more than 500,000 people die every year from antibiotic resistance, about 30,000 of them due to CPE. The AI analyzes inhibitory zones in the antibiogram – diameters of growth inhibitory zones around antibiotic platelets – and outperforms algorithms such as EUCAST or the French Society of Microbiology. While these 27.8 to 61 percent deliver false-positive hits, the rate with CarbaDetector is only 13 percent, with consistently high sensitivity.

Currently only approved for research purposes in Germany, CarbaDetector is to be further developed and made available free of charge to resource-poor countries. This could significantly support the global fight against multi-resistant germs.

Original Paper:

CarbaDetector: a machine learning model for detecting carbapenemase-producing Enterobacterales from disk diffusion tests | Nature Communications

Editor: X-Press Journalistenbüro GbR

Gender Notice. The personal designations used in this text always refer equally to female, male and diverse persons. Double/triple naming and gendered designations are used for better readability. ected.