Using neural networks for pancreatic cancer screening could increase the number of cases detected from 10% to 35%. This has been shown by a study published in the journal. eBioMedicine.
Pancreatic ductal adenocarcinoma is a dangerous disease with very low survival rates. Early detection of a tumor can increase the chances of cure, but still 80-85% are diagnosed in late stages when treatment is no longer possible.
In the new study, researchers used data from more than 5 million patients at institutions across the United States. The combination of PRISM neural network and logistic regression model for analysis outperformed existing methods. Standard screening criteria detect approximately 10% of adenocarcinomas, but PRISM can detect 35% of tumors.
The models analyzed up to 85 patient variables to predict cancer risk: demographics, diagnoses, medications and laboratory results.
“The hypothesis was that medical records contained hidden clues—subtle signs and symptoms that may indicate an increased risk of pancreatic cancer,” the scientists explained.
The study may lead to the development of new strategies to identify patients at increased risk of malignancy who may benefit from more comprehensive screening.
previously gynecologist warned Women over 35 about dangerous “silent” cancer.