Pancreas cancer segmentation on portal-venous phase CTs

This dataset consists of reference segmentations for 91 abdominal CTs delineating the pancreas and pancreas adenocarcinoma (PDA). The CTs were downloaded from freely available public archives for pancreas AI applications- The Cancer Imaging Archive (TCIA) CPTAC-PDA (n=60) and Medical Segmentation Decathlon (n=420) and evaluated by two radiologists for image quality, phase of image acquisition, presence of biliary stents and etiology of the pancreatic lesion. Volumetric pancreas and PDA segmentations was done on the portal-venous phase CTs on 3D Slicer software, by a radiologist with 4-year post residency experience. The PDA segmentations were subsequently verified by an expert abdominal radiologist with 11-years post residency experience. Segmentations were performed only on CTs that did not have expert segmentations as a part of original dataset.

For the CPTAC-PDA dataset, segmentations were performed on 42 out of 60 CTs after excluding 18 CTs for the following reasons: presence of biliary stent (n=10), missing slices through pancreas (n=4), post-pancreatectomy status (n=1), incorrect region of scan (n=1) and unavailable portal-venous phase (n=2). 

For the MSD dataset, segmentations were performed on 49 out of 139 ‘testing’ subset CTs after excluding 90 CTs for the following reasons: presence of biliary stent (n=38), non-PDA pathology [n=36 for IPMN, n=14 for PNET) and post-treatment status (n=2). 

The uploaded files are in NIfTI format. In addition, a metadata has been uploaded containing the following information for each segmented CT: location of tumor in pancreas, T-stage, single largest diameter of tumor, presence or absence of pancreatic atrophy, dilatation of main pancreatic duct and common bile duct. 

This dataset has been referenced and elaborated in the paper-https://pubmed.ncbi.nlm.nih.gov/33840636/.