NVIDIA collaborates with a dozen international institutions on a computed tomography (CT) dataset for pancreatic research

The Pancreatic Tumor Segmentation (PanTS) dataset, a collaborative effort between 12 global institutions and NVIDIA, has been created to improve early-detection rates for pancreatic cancer. This extensive dataset, which includes 36,000 scans from 145 institutions worldwide, is aimed at revolutionising the early detection of this deadly disease.

The PanTS dataset boasts expert-validated annotations for over 993,000 structures, including tumors, regions of the pancreas, and 24 surrounding anatomical structures such as vascular, skeletal, abdominal, and thoracic structures. The inclusion of these additional structures is believed to contribute significantly to the improved performance of AI models trained on the PanTS dataset.

One of the key factors that set the PanTS dataset apart is its 16x larger-scale tumor annotations, which have been attributed to the improved performance of AI models trained on it. These models have demonstrated significantly better results than those trained on existing public datasets.

Pancreatic cancer is a formidable foe, with approximately 80% to 85% of cases being detected too late for effective treatment. The early detection offered by the PanTS dataset could potentially save countless lives. In the United States, pancreatic cancer is the third-leading cause of cancer deaths.

To facilitate the use of the PanTS dataset, NVIDIA has employed NVIDIA's open-source medical imaging AI framework, MONAI, for pancreatic tumor detection, localization, and segmentation. The dataset and the MONAI framework together offer a powerful tool in the fight against pancreatic cancer.

While the 12 collaborating institutions are not individually named in the available search results, the collaboration represents a significant stride towards improving the early detection of pancreatic cancer. The PanTS dataset is a testament to the power of collaboration and the potential of AI in the medical field.

Access to the PanTS dataset can be found here. With its extensive annotations and collaborative approach, the PanTS dataset offers a promising future in the early detection and treatment of pancreatic cancer.