Artificial intelligence identifies bacteria images quickly, accurately
Researchers at Beth Israel Deaconess Medical Center are using artificial intelligence to identify images of bacteria quickly and accurately through an AI-enhanced microscope, which they contend has the potential to alleviate the current national shortage of clinical microbiologists.
BIDMC’s Clinical Microbiology Laboratory is a “hidden part” of the Boston hospital, explains James Kirby, MD, director of the lab, but one that serves a critical function in diagnosing potentially deadly blood infections which is passed along to clinicians to determine appropriate therapies.
“We have a microbiology technologist workforce, and one the things they spend a lot of time doing is looking at patient specimens in order to make a diagnosis of the type of infection people have,” says Kirby, who is also associate professor of pathology at Harvard Medical School. “That’s a very labor-intensive task. It takes time, and it takes a lot of skill.”
However, Kirby notes that there is a nationwide shortage of highly trained microbiologists, with 9 percent of lab technologists positions remaining unfilled—a situation that will only get worse with 20 percent of technologists projected to reach retirement age during the next five years, according to the American Society for Clinical Pathology.
To fill this gap, BIDMC researchers are leveraging an automated AI-enhanced microscope system from MetaSystems with a digital camera—which collects high-resolution image data—and so far has proven “highly adept” at identifying images of bacteria quickly and accurately based on their shape and distribution, according to Kirby.
Researchers specifically trained a class of AI modeled on the mammalian visual cortex, called convolutional neural network (CNN), to analyze visual data from blood samples on microscopic slides in order to categorize bacteria in suspected bloodstream infections.
“One characteristic of these convolutional neural networks is the more you train it, the better it becomes” at these tasks, observes Kirby, who says his research team generated more than 100,000 training images for the system. “It’s like a child going from primary to secondary school.”
Results of a study, published recently in the Journal of Clinical Microbiology, show that the CNN was able to sort images into the three categories of bacteria—rod-shaped, round clusters and round chains or pairs—ultimately achieving nearly 95 percent accuracy.
In addition, researchers had the algorithm sort new images from 189 slides without human intervention, which achieved more than 93 percent accuracy in all three bacteria categories.
“We view this as a first step as a proof of principle,” says Kirby, who believes that in the future with further development and training the AI-powered platform could be used as fully automated bacteria classification system, “conceivably reducing technologist read time from minutes to seconds.”
Time is of the essence for clinical microbiologists, he notes, as rapid identification and delivery of antibiotic medications is the key to treating bloodstream infections, which kill as many as 40 percent of patients who develop them.
“You have the power of the machine’s intelligence and the power of the microbiologist together to render a diagnosis,” Kirby concludes. “I work at a big teaching hospital, but there are many smaller sites that don’t have local expertise. Slides could be scanned at these remote sites and then analyzed potentially at a central site to suggest what the diagnosis might be.”
Besides clinical applications, the AI-based system could also be utilized for microbiology training and research as a “living data repository,” he adds.