Reproduce “Early and Prediagnostic Detection of Pancreatic Cancer from Computed Tomography” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wenxuan Li, Pedro R. A. S. Bassi, Lizhou Wu, Xinze Zhou, Yuxuan Zhao, Qi Chen, Szymon Plotka, Tianyu Lin and 29 more
ePAI is a three-stage CT analysis system designed to detect and anatomically localize pancreatic ductal adenocarcinoma, including tumors measuring 2 cm or less and abnormalities on scans acquired before diagnosis. Across 7,158 external cases, it reached a pooled AUC of 0.971, 97.0% sensitivity, and 88.0% specificity; in a retrospective prediagnostic cohort, it identified cancer directly in 47% of patients a median of 347 days before clinical diagnosis.
Key insights
- Anatomical context and voxel-level predictions allow ePAI to detect and localize disease; localization was 94.6% for internal all-size tumors and 88.7% on the external cohort, but dropped to 79.7% for small external tumors.
- Prediagnostic recognition was not limited to visible masses: among cases without direct tumor detection, the model identified duct dilation or related morphological abnormalities in 85%, indicating a possible route to earlier clinical review.
- The reader study supports an assistive role: ePAI substantially outperformed unaided readers on early and prediagnostic cases, while providing additional sensitivity gains when radiologists reviewed its outputs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.