Running title: CT-based subtypes of pancreatic ductal adenocarcinoma. Keywords Statement of translational relevanceThere are currently limited methods to stratify patients with pancreatic ductal adenocarcinoma (PDAC) into prognostic groups based on disease biology. We focused on the morphological features of the disease on computed tomography (CT) scans and correlated these features with genomic, pathological, and clinical data. We found that tumors with a distinct interface in relation to the surrounding parenchyma (called high delta tumors) had more aggressive biological features than tumors without a distinct interface (called low delta tumors). Patients with high delta tumors were more likely to develop early distant metastasis and die faster than those with low delta tumors. Mathematical modeling suggested that stromal elements strongly influenced the morphological patterns seen on CT scans. These findings indicate that a universally available diagnostic test can be used to interrogate the biology of this deadly disease, providing a means to stratify patients at diagnosis and aiding the design of future clinical trials.

<p>Mathematical model simulation of high delta tumor growth (lambda = 1.5)</p>

<p>Mathematical model simulation of low delta tumor growth (lambda = 0.2)</p>

<p>Additional methods</p>

<p>Includes Supplementary Tables S1 through S9, and Supplementary Figures S1 through S12</p>