Researchers at the Institute of Clinical and Preventive Medicine at the University of Latvia and the Institute of Electronics and Computer Science want to build a tool for doctors - an artificial intelligence-based diagnostic assistant, AIDA for short - to help detect stomach cancer. To teach the virtual mind to recognize damaged cells, it first had to build up a database of images of tumors in different states. The microscopy image of the tissue sample is converted into a digital file so that it can be run in computer programs.
"The samples are scanned, we open them and mark the regions we are interested in, to distinguish where there is a tumor and where there is not," explained Jānis Ziemelis, pathologist-in-residence at the Rīga East Clinical University Hospital. "Our job as pathologists is to map the structure of a gastric tumor. We point out where we see tumor structures and where we see an unchanged layer of the stomach wall."
The pathologists' images are given to artificial intelligence to analyze. It compares the images where there is a tumor and where there is not, and looks for pathology matches in new samples.
"It's like a student, the more samples it sees, the better it will be able to recognize other similar images," explained Institute of Electronics and Computer Science researcher Arturs Nikulins.
The result produced by the artificial intelligence shows where the cancer cells are located in green. Inese Poļaka, a senior researcher at the Institute of Clinical and Preventive Medicine, highlights the good results of AI training.
"The model has shown quite good accuracy. 92% of the time it correctly recognises where it has shown these cells. We will further add other types of cancer and continue to train it so that it can also recognize precancerous conditions and help in diagnosis," said Poļaka. "In regions where histopathologists may not be as experienced, mistakes can happen more often, so we have artificial intelligence that could indicate - pay attention here - or help train pathologists."
Malignant tumors are the second most common cause of death in Latvia after diseases of the circulatory system. Oncology in Latvia has one of the highest mortality rates in the European Union.
"Artificial intelligence will never replace pathologists, but if something can help us in our daily lives, why not?" added Ziemelis.
Work on an AI tool for cancer diagnosis is only halfway through. This is an international project, with France, Spain, Portugal, and other countries participating.
