Massachusetts Institute of Technology (MIT) announced the launch of two new products called "PRISM"ofArtificial Intelligence Model, designed to detect pancreatic cancer earlier than traditional methods.
Source Note: The image is generated by AI, and the image is authorized by Midjourney
Pancreatic cancer is a dangerous disease. The early detection rate of traditional diagnostic methods is only about 10%, which greatly delays the treatment opportunity. The PRISM system significantly improves this rate.
Standard screening criteria only screen out about 1 in 10% cases of pancreatic ductal adenocarcinoma (PDAC), and the relative risk threshold needs to be increased fivefold to achieve this. The PRISM model can identify 35% PDAC cases at the same threshold. Early detection of pancreatic cancer is crucial because timely treatment can save lives.
The researchers developed two models: a PRISM neural network and a logistic regression model. Both models analyzed electronic medical records, including patient demographics, diagnoses, medications, and laboratory results, to assess PDAC risk.
The PRISM model uses a neural network to identify complex patterns in these data points and calculates a risk score that estimates the likelihood of PDAC. The logistic regression model uses a simpler analytical method to generate a PDAC probability score based on these features.
All in all, PRISM was trained on data from more than 5 million patient records. This massive dataset enables the algorithm to identify patterns that human doctors might miss.
MIT has extensive experience developing AI models for cancer diagnosis, such as predicting breast cancer risk. These projects have shown that the greater the diversity of the dataset, the higher the accuracy of the diagnosis.
The PRISM project began six years ago with the goal of improving the detection of pancreatic cancer, and currently 80%'s pancreatic cancer patients are diagnosed at an advanced stage.
Although the PRISM model is promising, it still needs further development. Currently, the model is based only on U.S. data and needs to be tested and adapted before it can be used globally.
In the future, the research team plans to expand the model's application to international datasets and integrate more biomarkers for more refined risk assessment. The researchers also hope to promote the implementation of the model in routine healthcare.
Their vision is for the model to fit seamlessly into the backend of healthcare systems, automatically analyzing patient data and alerting doctors to high-risk cases without adding to their workload.
As the researchers write: "While the PRISM model is promising, as with all research, some aspects of it remain a work in progress."