Summary:
MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) has developed PRISM, two machine learning models – PrismNN and PrismLR – that offer hope for early identification of high-risk individuals in pancreatic cancer. These models outperform current screening methods by detecting 35% of pancreatic ductal adenocarcinoma (PDAC) cases at the same risk threshold, compared to the existing 10%.
PRISM’s strength lies in its vast training data, sourced from anonymized electronic health records of over 5 million U.S. patients. The models learn subtle patterns and generalizable risk factors, overcoming geographic constraints. Transparency is a priority, with around 85 key indicators highlighted, aligning with known risk factors for pancreatic cancer.
The models currently operate on U.S. data, but global adaptation, further testing, and refinement are underway. Future goals include integrating additional biomarkers and seamless integration into healthcare systems.
The vision for PRISM is to silently alert physicians to high-risk cases, enabling interventions before symptoms manifest. Harvard Medical School professor David Avigan sees great promise in this approach, potentially leading to earlier interventions and improved outcomes.
Published in the open-access journal eBioMedicine, this research marks a significant stride in early pancreatic cancer detection, offering hope to patients and families. With continued development, PRISM could become a powerful tool in the fight against this challenging disease.
Article written by Charles Sternberg
22/01/2024
Source:
MPO MAG
https://www.mpo-mag.com/contents/view_breaking-news/2024-01-22/researchers-detect-pancreatic-cancer-with-machine-learning-models/