Researchers at the University of Cambridge have demonstrated that AI can be used to identify drug-resistant diseases, significantly reducing the time it takes to make an accurate diagnosis. Using a trained algorithm, the research group showed that it is possible to accurately identify drug-resistant bacteria from microscopic images alone. The study Nature Communications.
Antibiotic resistance is a growing health concern worldwide, making many infections harder to treat and reducing the number of available treatments, and may even make some infections untreatable in the near future.
Being able to quickly distinguish between microorganisms that are resistant to treatment and those that can be treated with first-line drugs is one of the challenges facing healthcare professionals.
Traditional testing involves culturing the bacteria, testing it against various antibacterial treatments, and then having a lab technician or machine analyze the results — a process that can take days. As a result of this delay, patients are often treated with the wrong drug, which can lead to more dire outcomes and increased drug resistance.
Researchers from Professor Stephen Baker's lab at the University of Cambridge led the team that developed the machine learning tools that can make the recognition possible. Salmonella typhi Bacteria resistant to ciprofloxacin, a first-line antibiotic, were detected in the microscopic images without being tested for the drug.
In extreme cases, S. Typhi It can cause typhoid-like illness and gastrointestinal problems, with symptoms including fever, headache, nausea, fatigue, stomach discomfort, constipation and diarrhea. In extreme cases, it can be fatal. Antibiotics can cure the disease, but it is becoming more difficult to treat as bacteria develop resistance to antibiotics.
The researchers S. Typhi Using high-resolution microscopy, we observed isolates exposed to increasing doses of ciprofloxacin and determined the five most important image characteristics for distinguishing resistant from susceptible isolates.
We then built and evaluated machine learning algorithms to detect these traits using image data from 16 samples.
Without exposing the bacteria to the drug, the algorithm was able to accurately predict in every case whether the bacteria were susceptible or resistant to ciprofloxacin – and this was for isolates that were cultured for just six hours, rather than the standard 24-hour incubation period in the presence of the antibiotic.
S. Typhimurium bacteria that are resistant to ciprofloxacin have several notable differences compared to bacteria that are still susceptible to this antibiotic, and although a skilled human operator may be able to identify some of these differences, this alone is not sufficient to reliably distinguish between resistant and susceptible bacteria. The beauty of machine learning models is that they can identify resistant bacteria based on some subtle features in microscopic images that the human eye cannot detect..
Dr Tuan Anh Tran of Cambridge University
At the time he worked on this research, Tran was a doctoral student at Oxford University.
To use this method to analyze samples such as blood, urine or stool, the bacteria must be isolated, but because the bacteria do not need to be tested for ciprofloxacin, the overall procedure could be reduced from days to hours.
The researchers argue that this particular strategy, despite its practical and cost-effectiveness limitations, shows how powerful artificial intelligence can be in the fight against antibiotic resistance.
This approach is not yet a solution that can be easily deployed everywhere, as it uses single-cell resolution images, but it shows great promise in that capturing just a few parameters about bacterial shape and structure provides enough information to predict drug resistance relatively easily..
Dr. Sushmita Sridhar, Postdoctoral Research Fellow, University of New Mexico
Dr. Sridhar, who is also affiliated with the Harvard School of Public Health, launched the project while he was a doctoral student at Cambridge Medical School.
The research team is currently working on a larger collection of bacteria to build a more reliable set of experiments and further accelerate the identification process, which will allow them to detect antibiotic resistance in a range of bacterial species, including ciprofloxacin resistance.
What's really important, especially in a clinical setting, is to be able to take complex samples like blood, urine, sputum, and directly identify susceptibility and resistance from them. This is a much more complex problem, and even clinical diagnostics in hospitals has not been solved at all. If we can find a way to make this happen, we can reduce the time it takes to identify drug resistance and reduce the cost significantly. This could be really transformative..
Dr. Sushmita Sridhar, Postdoctoral Research Fellow, University of New Mexico
Wellcome funded the research.
Journal References:
Tran, T.-A. others(2024) Combining machine learning and high-content imaging to infer ciprofloxacin susceptibility in Salmonella typhimurium isolates. Nature Communications. doi.org/10.1038/s41467-024-49433-4.
