MIT AI models detect the most common type of pancreatic cancer earlier

MIT CSAIL researchers have created two advanced artificial intelligence programs to help detect earlier. With a combination of a neural network (PRISMNN) and a logistic regression model (PRISMLR), they can find pancreatic ductal adenocarcinoma (PDAC) — the most common type of pancreatic cancer. The researchers, along with Dr Limor Apelbaum, a Harvard Medical School instructor as well as radiation oncologist, wanted to diagnose pancreatic cancer, which is usually hard to find in its early stages.

AI programs performed better than existing methods “Approximately 80 85 percent of pancreatic cancer patients are diagnosed at advanced stages, where cure is no longer an option,” said Appelbaum. “This clinical frustration sparked the idea to delve into the wealth of data available in electronic health records (EHRs).” They created two computer programs, called the “PRISM” neural network and a logistic regression model, to help identify high risk patients.

These programs performed better than current methods. The team used a big database of over five million patients from different healthcare institutions in the U.S. to make sure the programs would work well for various groups of people. PRISMNN uses artificial neural networks to find complex patterns in the data and gives a risk score for the likelihood of .

PRISMLR uses logistic regression for a simpler analysis, creating a probability score based on the same information. These models together provide a thorough evaluation of different ways to predict pancreatic cancer risk using the same EHR data. “A machine learning model integrated with the EHR system could empower physicians with early alerts for high risk patients, potentially enabling interventions well before symptoms manifest,” said Kai Jia, MIT CSAIL PhD student and senior author on the paper.

These models looked at details like patient information, diagnoses, medications, and lab results. When they compared the new programs to standard screening methods, the PRISM model could detect 35 percent of pancreatic cancer cases, while the usual methods could only find 10 percent. “A subsequent aim for us is to facilitate the models' implementation in routine healthcare settings.

The vision is to have these models function seamlessly in the background of healthcare systems, automatically analyzing patient data and alerting physicians to high risk cases without adding to their workload,” added Jia. Jia, along with Appelbaum and MIT professor Martin Rinard, worked on this paper.

They got help from various organizations like DARPA, Boeing, the National Science Foundation, and Aarno Labs while they were at MIT CSAIL. TriNetX also pitched in with resources for the project, and the Prevent Cancer Foundation supported the team. The was published in ..