AI-Powered Diagnostic Method Detects Sepsis in Two Hours, Revolutionizing Treatment

New method detects sepsis infections in just two hours

KTH The Royal Institute of TechnologyAug 28 2025 A new diagnostic method would confirm sepsis infections earlier, cutting critical hours in the "race against time" to save patients' lives. Publishing in Nature Publishing Journal Digital Medicine, the team from KTH Royal Institute of Technology and Uppsala University say their diagnostic process offers a speedier alternative to the bacteria culturing process hospitals routinely use to identify suspected bloodstream infections. The process uses a centrifuge to separate bacteria from blood cells, and automatic microscopy for detection, enabling a clinic to confirm bacterial infection in as little as two hours using software trained by artificial intelligence, says Henar Marino Miguelez, a doctoral student at KTH Royal Institute of Technology. She and doctoral student Mohammad Osaid were the study's lead authors. By contrast, hospital labs generally need at least a day of incubation before the growth of infectious bacteria begins to reveal itself in blood cultures. Diagnosing sepsis is a race against time. With every hour of delayed treatment of patients in septic shock, survival rates drop by 8 percent." Henar Marino Miguelez, doctoral student, KTH Royal Institute of Technology By enabling prompt identification of pathogens, the appropriate antibiotic treatment can be started sooner, says Wouter van der Wijngaart, a professor at KTH Royal Institute of Technology who leads research in microfluidic and biomedical systems. Typically a clinic will put a patient on a broad-spectrum antibiotic when sepsis is suspected, at least until they identify the pathogen. But that precaution carries its own risks, due to the inherent drug toxicity, attacking beneficial gut bacteria and promoting the emergence of antibiotic-resistant strains. "It takes a hospital two to four days before they are sure which antibiotic to treat a bloodstream infection with," van der Wijngaart says. "We're trying to do this in four to six hours." In tests using blood samples spiked with bacteria, the system successfully detected E. coli, K. pneumoniae, and E. faecalis at clinically relevant levels, as low as 7 to 32 bacterial colony-forming units per milliliter of blood. While the method proved to work well with these bacteria, it did not for staphylococcus aureus, which hides in blood clots. Miguelez says the researchers are working on ways to fix that. The technique employs a "smart centrifugation", which spins blood samples on top of an agent that causes bacteria to float upwards while blood cells sediment downwards, leading to a clear, liquid layer containing bacteria but no blood cells. This liquid is then injected into a chip with microscale channels, where it flows easily. Miniscule traps in the chip capture the separated bacteria, and any bacteria growth quickly becomes visible in automated time-lapse microscopy images analyzed by the machine learning software. The work was a collaboration between the teams of van der Wijngaart at KTH, and Johan Elf and Carolina Wählby at Uppsala University. KTH The Royal Institute of Technology Journal reference: Marino Miguélez, M. H., et al. (2025). Culture-free detection of bacteria from blood for rapid sepsis diagnosis. Npj Digital Medicine. doi.org/10.1038/s41746-025-01948-w

Faster diagnostic method can detect sepsis in hours instead of days

A new diagnostic method would confirm sepsis infections earlier, cutting critical hours in the "race against time" to save patients' lives. Publishing in npj Digital Medicine, the team from KTH Royal Institute of Technology and Uppsala University say their diagnostic process offers a speedier alternative to the bacteria culturing process hospitals routinely use to identify suspected bloodstream infections. The process uses a centrifuge to separate bacteria from blood cells, and automatic microscopy for detection, enabling a clinic to confirm bacterial infection in as little as two hours using software trained by artificial intelligence, says Henar Marino Miguelez, a doctoral student at KTH Royal Institute of Technology. She and doctoral student Mohammad Osaid were the study's lead authors. By contrast, hospital labs generally need at least a day of incubation before the growth of infectious bacteria begins to reveal itself in blood cultures. "Diagnosing sepsis is a race against time," Marino Miguelez says. "With every hour of delayed treatment of patients in septic shock, survival rates drop by 8%." By enabling prompt identification of pathogens, the appropriate antibiotic treatment can be started sooner, says Wouter van der Wijngaart, a professor at KTH Royal Institute of Technology who leads research in microfluidic and biomedical systems. Typically, a clinic will put a patient on a broad-spectrum antibiotic when sepsis is suspected, at least until they identify the pathogen. But that precaution carries its own risks, due to the inherent drug toxicity, attacking beneficial gut bacteria and promoting the emergence of antibiotic-resistant strains. "It takes a hospital two to four days before they are sure which antibiotic to treat a bloodstream infection with," van der Wijngaart says. "We're trying to do this in four to six hours." In tests using blood samples spiked with bacteria, the system successfully detected E. coli, K. pneumoniae, and E. faecalis at clinically relevant levels, as low as seven to 32 bacterial colony-forming units per milliliter of blood. While the method proved to work well with these bacteria, it did not for staphylococcus aureus, which hides in blood clots. Miguelez says the researchers are working on ways to fix that. The technique employs a "smart centrifugation," which spins blood samples on top of an agent that causes bacteria to float upwards while blood cells sediment downwards, leading to a clear, liquid layer containing bacteria but no blood cells. This liquid is then injected into a chip with microscale channels, where it flows easily. Minuscule traps in the chip capture the separated bacteria, and any bacteria growth quickly becomes visible in automated time-lapse microscopy images analyzed by the machine learning software. The work was a collaboration between the teams of van der Wijngaart at KTH, and Johan Elf and Carolina Wählby at Uppsala University.