A study has found that pancreatic cancer, which is assessed as an incurable cancer, may be predicted early with artificial intelligence (AI). Pancreatic cancer has the bottom 5-year survival rate because early detection is difficult.

The Washington Post reported on the seventeenth (local time) that researchers from the University of Copenhagen in Denmark and Harvard Medical School in america built an AI model that may discover high-risk patients with pancreatic cancer based on data from a particular patient group, and published a paper containing the research in a medical journal. reported in Nature Medicine.

In response to the report, the researchers used data from the medical records of 6.2 million patients in Denmark and three million US veterans between 1977 and 2020.

When a machine learning model was trained to predict pancreatic cancer risk based on a patient’s symptoms and various diagnostic codes of their medical records, the AI ​​model identified aspects that increased risk up to a few years prior to being diagnosed with pancreatic cancer.

The chance aspects identified by AI included symptoms not traditionally related to pancreatic cancer, akin to gallstones, type 2 diabetes, anemia, vomiting and abdominal pain.

In fact, the researchers explained that these symptoms cannot necessarily be thought to be precursors or causes of pancreatic cancer, and that risk aspects may vary depending on the characteristics of the dataset. With the intention to construct an AI model, it should be trained to reflect the characteristics of the local population.

In a press release related to the study, Harvard Medical School explained that the advantage of the AI ​​tool is that it could be used not just for patients with a known family history or genetic predisposition, but in addition for all patients who’ve access to data on their health history and other medical history. did.

Because high-risk patients may not know their genetic predisposition or family history, using these AI tools could enable early diagnosis through higher screening or targeted testing.

Reporter Jeong Byeong-il jbi@aitimes.com