AI has learned to detect cancer long before symptoms appear
Scientists have developed an artificial intelligence system capable of detecting pancreatic cancer up to a year and a half before an official diagnosis, effectively “spotting” the disease where doctors cannot yet see it. In tests, the model demonstrated greater accuracy than radiologists and was able to identify early signs of the disease even in the absence of obvious symptoms, according to ScienceAlert.
Researchers from the Mayo Clinic and the University of Texas Cancer Center have developed the REDMOD artificial intelligence model, which analyzes CT scans and detects subtle changes in pancreatic tissue that may indicate the development of cancer long before clinical signs appear.
“We aren’t looking for the tumor itself; we’re looking for subtle changes in tissue structure that the human eye might miss,” the researchers explain, noting that the algorithm works by analyzing radiomic patterns—that is, microchanges in the organ’s texture.
The system was trained on hundreds of CT scans and then tested on individual patient data from those who were later diagnosed with cancer. It turned out that in nearly three out of four cases, REDMOD correctly identified the risk even before the diagnosis was made, whereas doctors without AI assistance demonstrated significantly lower accuracy. In some cases, the model “detected” suspicious changes more than two years before diagnosis, which, according to the authors, potentially paves the way for even earlier detection of the disease—sometimes as much as three years before it is confirmed.
However, for some healthy patients, the system also generated false alarms, mistakenly flagging images as suspicious, which means additional testing and further refinement of the technology are needed before it can be used in clinics. The researchers emphasize that the technology’s greatest potential lies in analyzing routine CT scans performed for other reasons, as this is precisely how cancer can be incidentally detected in its early stages, when treatment is still most effective.
The system is currently being tested on larger patient samples to determine how it can be integrated into real-world medical practice, and the study results have already been published in the journal Gut.
Researchers at Argonne National Laboratory in the U.S. are developing robotic assistants capable of teaching themselves how to perform complex experiments simply by observing human actions.
Japan Airlines plans to launch a pilot program in May to use humanoid robots for baggage handling at Tokyo’s Haneda Airport. This is intended to partially offset staff shortages and reduce the workload on employees.
In Germany, there are plans to ban nighttime operation of robotic lawn mowers due to hedgehogs.
