The threat of antibiotic-resistant bacteria is continually increasing. Rapid, on-the-spot diagnosis of microbial infections using a smartphone could result in more judicious prescription of antibiotics.© AnaLysiSStudiO/Shutterstock
Unchecked, the number of global deaths from antimicrobial resistance could rise to 10 million annually by 2050, according to a report by the World Health Organization. New image-recognition technology aims to curb the excessive use of antibiotics by helping diagnose bacterial infections using smartphones within seconds. Developed by CarbGeM Inc., a spin-off company of Japanese bioventure NextGeM Inc., the technology uses deep learning to identify bacteria from Gram stains, a common test used to classify bacteria into two broad categories based on their cell-wall characteristics.
A major factor in the emergence of antibiotic-resistant bacteria is the wide use of broad-spectrum antibiotics targeting various types of bacteria. Isao Miyatsuka, director of CarbGeM, hopes that image recognition will help curb antimicrobial resistance by matching infections to narrow-spectrum antibiotics, which are extremely effective for specific types of bacteria.
“Physicians are often pressured to prescribe something immediately to treat patients, even though it is difficult to determine infection-causing bacteria during the early stages of an infection. Results from conventional culture techniques take one to three days to obtain,” Miyatsuka explains. “Identifying bacteria types is easy if specialized analysis equipment or PCR facilities are available, but this is unrealistic for most physicians in primary care settings.”
CarbGeM is starting with bacteria from urine samples, in partnership with the National Center for Global Health and Medicine (NCGM) and Kobe University. The technology identifies 20 major types of bacteria — covering more than 90% of the infection-causing bacteria found in patients. “Gram stains show whether the bacteria in urine samples are gram positive or negative by staining them purple or pink,” says Miyatsuka. “Only the most experienced technicians can make further distinctions by examining the bacteria’s form through microscopes — the accuracy of our image-recognition technology now matches that level of accuracy.”
The Gram stains are captured using an attachment to a smartphone camera, which connects to the eyepiece of optical microscopes. “Key opinion leaders suggested we develop the technology on smartphones, as physicians in many parts of the world have them at hand,” says Miyatsuka.
A screenshot of the smartphone app that uses artificial intelligence to rapidly identify bacterial infections.© CarbGeM Inc.
An artificial-intelligence model on the cloud generates results in roughly 30 seconds, including a degree of confidence. The system suggests the most promising antibiotics based on antibiograms, the collection of data about the susceptibility of bacteria in each facility or region to various antibiotics.
Researchers at CarbGeM are continuing to improve the model. “With widespread use of this system, we expect a better accuracy as the system learns from accumulated data on more bacterial species and samples,” says Miyatsuka.