Researchers develop AI-powered method to diagnose lung infections faster

Researchers develop AI-powered method to diagnose lung infections faster

Lung infections such as pneumonia are among the leading causes of death worldwide - but they are notoriously difficult to diagnose.

Now researchers at UC San Francisco have found a way to identify these infections in critically ill patients by combining generative AI analysis of medical records with a biomarker for lower respiratory tract infections.

In an observational study of critically ill adults, the combination made a correct diagnosis 96 percent of the time and differentiated between infectious and non-infectious causes of respiratory failure more accurately than ICU doctors. The authors estimated that this model, if available when patients were enrolled, could have reduced inappropriate antibiotic use by more than 80%.

“We have developed a method that produces results much more quickly than culture, and it could be easily implemented in the clinic,” said Dr. Chaz Langelier, associate professor of medicine and senior author of the study published Dec. 16Nature communication.

We are confident that this could lead to faster diagnosis and reduce unnecessary use of antibiotics.”

Chaz Langelier, Associate Professor of Medicine, University of California – San Francisco

An important feature of the model is the biomarker that Langelier's team developed in 2023. They found that it is a gene that modulates inflammation, calledFABP4could help diagnose infection as it is less expressed in immune cells compared to normal lung cells.

The current study examined data from two groups of critically ill patients: 98 were recruited before the COVID-19 pandemic and most had bacterial infections; 59 were recruited during the pandemic and most had viral infections, including COVID-19.

First they tested each method individually –FABP4Biomarkers or AI – and found that each of them made the correct diagnosis about 80% of the time. The researchers then compared the model's results with the diagnoses of the doctors who admitted the patients to the hospital's intensive care unit.

These doctors prescribed antibiotics to treat pneumonia to most of these patients, while the biomarker-plus-AI model was much more reasonable in diagnosing pneumonia.

To further test the model's accuracy, the team compared the way the AI ​​analyzed the medical records with the way three different doctors specializing in internal medicine and infectious diseases analyzed them. The AI ​​was performed by GPT4 on a privacy platform developed at UCSF.

Both received roughly the same number of correct diagnoses, but the AI ​​placed more emphasis on radiology reports on chest X-rays while the doctors focused on clinical notes.

“It almost showed a cultural difference if you can say that about an AI,” said Natasha Spottiswoode, MD, DPhil, assistant professor of medicine, one of the paper’s first authors. “It shows how AI can complement the work of doctors.”

The team published their AI prompts in the paper and encouraged physicians to try them out on their own HIPAA-compliant AI platforms.

“It's incredibly easy to use, you don't have to be a bioinformatician,” said Hoang Van Phan, PhD, a bioinformatician himself and first author of the paper.

The team is validating the model as a clinical test. Next, they turn to sepsis, the leading cause of hospital death, which is also notoriously difficult to determine.

Sources: