NVIDIA has collaborated with 12 global institutions to create a comprehensive pancreatic CT dataset, with the aim of improving early-detection rates for pancreatic cancer.
The Pancreatic Tumor Segmentation dataset (PanTS) includes 36,000 scans from 145 institutions, with expert-validated annotations for over 993,000 structures, including tumors, regions of the pancreas, and 24 surrounding anatomical structures (e.g., vascular and skeletal structures, abdominal and thoracic organs).
The PanTS dataset uses NVIDIA’s open-source medical imaging AI framework, MONAI, for pancreatic tumor detection, localization, and segmentation. In analysis, AI models trained on PanTS have performed significantly better than models trained on existing public datasets, findings the developers say are attributable to the PanTS dataset’s 16x larger-scale tumor annotations, supported by the inclusion of the 24 additional surrounding anatomical structures.
Pancreatic cancer is the third-leading cause of U.S. cancer deaths; 80% to 85% of cases are detected too late for effective treatment.
Read about PanTS and get access to the dataset here.
