Latvian researchers are working on artificial intelligence that will recognize cancer cells, with an accuracy of 92%. The project is undertaken by representatives of the University of Latvia’s Institute of Clinical and Preventive Medicine and the Institute of Electronics and Computer Science – who are specifically developing a diagnostic tool for doctors in the case of stomach cancer.
The training of the AI model demands a large database of images of a tumor at different stages. The tissue samples are scanned and then transferred into digital format for analysis by AI programs. In this process, pathologists play an important role at Riga East Clinical University Hospital, marking areas of interest on the image and identifying regions that correspond to a tumor-a segment identified so that it is distinguished from healthy tissue.
According to Jānis Ziemelis of the hospital, who explained that mapping the structure of gastric tumors and showing an area where abnormal growth can be perceived is precisely what their duties mean. As such, the AI analyzes these marked images and learns to identify patterns by comparing tumor and non-tumor samples and with more improved data, the ability of AI to discover cancerous cells improves.
The AI output highlights potential locations of cancer in green, which serves as a basis in the diagnosis by a pathologist. According to Inese Poļaka, senior researcher at the Institute of Clinical and Preventive Medicine, “AI performance was impressive,” the AI caught cancer cells with 92% accuracy. She added that future developments will allow the AI to detect all types of cancers and various precancerous states.
Although AI is unlikely to replace pathologists, it certainly can assist them in minimal diagnosis at least where expert pathologists are scarce. It is part of an international cooperation project between researchers from France, Spain, Portugal, and other countries. As the cause of death number two in Latvia, cancer can benefit a great deal from this AI tool for early detection and results enhancement.
