Cambridge researchers have shown that artificial intelligence (AI) can quickly identify drug-resistant infections from microscope images, reducing diagnosis time. The AI can spot subtle features in images humans can’t see, quickly distinguishing resistant bacteria.
Antimicrobial resistance is a growing global health problem, making many infections hard to treat. One major challenge is identifying whether bacteria resist first-line drugs, as traditional testing can take days. This delay often leads to incorrect treatments, worsening patient outcomes and contributing to increased drug resistance.
Researchers from Professor Stephen Baker’s Lab at the University of Cambridge created a machine-learning tool to identify ciprofloxacin-resistant Salmonella Typhimurium from microscopy images without testing the bacteria against the drug.
S. Typhimurium can cause serious illness with symptoms like fever, headache, and abdominal pain. In severe cases, it can be life-threatening. While antibiotics can treat infections, the bacteria are becoming more resistant, making treatment more challenging.
Credit: Rocky Mountain Laboratories, NIAID, NIH
The team used high-resolution microscopy to study S. Typhimurium exposed to ciprofloxacin and identified the top five imaging features that indicate resistance. Using data from 16 samples, they trained a machine-learning algorithm to recognize these features.
The algorithm correctly predicted whether bacteria were resistant or susceptible to ciprofloxacin without exposing them to the drug. It worked on samples cultured for six hours, much faster than the usual 24 hours.
Dr. Tuan-Anh Tran, who worked on this research, explained that ciprofloxacin-resistant S. Typhimurium bacteria differ from susceptible ones in several ways. While experts might spot some differences, more is needed to be sure.
“The beauty of the machine learning model is that it can detect resistant bacteria based on subtle features in microscopy images that human eyes can’t see,” he said.
Bacteria must still be isolated from blood, urine, or stool samples to use this approach. However, since the bacteria don’t need to be tested against ciprofloxacin, the process could be shortened from several days to just a few hours.
While this method may have practical and cost limitations, it shows how powerful AI could be in fighting antimicrobial resistance.
Dr. Sushmita Sridhar, who started this project during her PhD and is now a postdoc, said, “This approach uses single-cell imaging and isn’t ready for widespread use yet. But it shows promise by capturing a few details about the bacteria’s shape and structure, helping us predict drug resistance easily.”
The team plans to work with more extensive collections of bacteria to create a more robust system, speeding up the identification process and detecting resistance to ciprofloxacin and other antibiotics in different bacteria species.
Sridhar said, “In a clinical setting, it would be important to identify susceptibility and resistance directly from complex samples like blood, urine, or sputum. This is a complicated problem that hasn’t been solved yet, even in hospitals. If we could achieve this, we could quickly identify drug resistance at a lower cost, which would be transformative.”
Journal reference:
- Tran, TA., Sridhar, S., Reece, S.T. et al. Combining machine learning with high-content imaging to infer ciprofloxacin susceptibility in isolates of Salmonella Typhimurium. Nature Communications. DOI: 10.1038/s41467-024-49433-4.