Navigation auf uzh.ch

Suche

UZH News

Microbiology

AI Helps to Detect Antibiotic Resistance

In a pilot study, researchers at the University of Zurich have used artificial intelligence to detect antibiotic resistance in bacteria for the first time. This is an important first step toward integrating GPT-4 into clinical diagnostics.

Kategorien

In the laboratory, bacteria are often cultivated in Petri dishes on a nutrient medium that supplies the growing microorganisms with water and nutrients. (Image: iStock.com/zmeel)
In the laboratory, bacteria are often cultivated in Petri dishes on a nutrient medium that supplies the growing microorganisms with water and nutrients. (Image: iStock.com/zmeel)

Researchers at the University of Zurich (UZH) have used artificial intelligence (AI) to help identify antibiotic-resistant bacteria. The team led by Adrian Egli, UZH professor at the Institute of Medical Microbiology, is the first to investigate how GPT-4, a powerful AI model developed by OpenAI, can be used to analyze antibiotic resistance.

The researchers used AI to interpret a common laboratory test known as the Kirby-Bauer disk diffusion test, which helps doctors to determine which antibiotics can or can’t fight a particular bacterial infection. Based on GPT-4, the scientists created the “EUCAST-GPT-expert”, which follows strict EUCAST (European Committee on Antimicrobial Susceptibility Testing) guidelines for interpreting antimicrobial resistance mechanisms. By incorporating the latest data and expert rules, the system was tested on hundreds of bacterial isolates, helping to identify resistance to life-saving antibiotics.

Human experts are more accurate – but AI is faster

“Antibiotic resistance is a growing threat worldwide, and we urgently need faster, more reliable tools to detect it,” says Adrian Egli, who led the study. “Our research is the first step toward using AI in routine diagnostics to help doctors identify resistant bacteria more quickly.”

The AI system performed well in detecting certain types of resistance, but it wasn’t perfect. While it was good at spotting bacteria resistant to certain antibiotics, it sometimes flagged bacteria as resistant when they were not, leading to possible delays in treatment. In comparison, human experts were more accurate in determining resistance, but the AI system could still help standardize and speed up the diagnostic process.

Useful tool to support medical staff

Kirby-Bauer disk diffusion test of gut bacteria: Paper sheets soaked with antibiotics are placed on a petri dish. The antibiotic concentration decreases with increasing distance. The closer bacteria grow to the test sheets, the more resistant they are (red circles). If the gradients of two different antibiotics meet, their effectiveness can increase (yellow arrows). (Image: UZH)

Despite the limitations, the study highlights the transformative potential of AI in healthcare. By offering a standardized approach to the interpretation of complex diagnostic tests, AI could eventually help reduce the variability and subjectivity that exists in manual readings, improving patient outcomes.

Adrian Egli emphasizes that more testing and improvements are needed before this AI tool can be used in hospitals. “Our study is an important first step, but we are far from replacing human expertise. Instead, we see AI as a complementary tool that can support microbiologists in their work,” he says.

Curbing the global development of antibiotic resistance

According to the study, AI has the potential to support the global response to antibiotic resistance development. With further development, AI-based diagnostics could help laboratories worldwide improve the speed and accuracy of detecting drug-resistant infections, helping to preserve the effectiveness of existing antibiotics.

 

Literature

Christian G. Giske, Michelle Bressan, Farah Fiechter, Vladimira Hinic, Stefano Mancini, Oliver Nolte, Adrian Egli. GPT-4 based AI agents – the new expert system for detection of antimicrobial resistance mechanisms? Journal of Clinical Microbiology. 17 October 2024. DOI: https://doi.org/10.1128/jcm.00689-24