The American College of Radiology (ACR) Data Science Institute (DSI) and the Cancer Imaging Archive (TCIA), funded by the National Cancer Institute (NCI), have teamed up to connect use cases and datasets to speed medical imaging artificial intelligence (AI) development. Dataset collection is the most important step in developing robust AI algorithms.
Linking ACR DSI Define-AI use cases to datasets enables developers to build radiology AI algorithms that include defined data elements and DICOM images useful for inputs, outputs, and training and testing models.
TCIA datasets have been matched to ACR DSI cancer and non-cancer use cases based upon attributes such as body area, modality, and presence of secondary comorbidities. TCIA data are available under Creative Commons Attribution Licenses, and most are freely available for commercial use for machine learning purposes.
“Finding a good source of data and adequate data for model testing and validation is an ongoing challenge for those developing AI. We’ve now made it easier for those developing AI for medical imaging to get good data, and we’re sharing it along with the guidelines provided by our use cases to be sure the algorithms developed have value for us as radiologists,” says Bibb Allen Jr., MD, FACR, ACR DSI chief medical officer and diagnostic radiologist at Grandview Medical Center in Birmingham, Ala.
“We are pleased to see the ACR referring investigators to the TCIA data resource to help tackle the challenges of training AI models for specific clinical use cases. The National Cancer Institute (NCI) Cancer Imaging Program (CIP) mission includes supporting the availability of public data for imaging research,” says Janet F. Eary, MD, associate director of the Cancer Imaging Program at the National Institutes of Health National Cancer Institute.
ACR DSI Define-AI use cases were linked to TCIA datasets based on ongoing requests from the developer community to improve access to data for algorithm development. Connecting Define-AI use cases with TCIA datasets may advance research and development by providing AI developers with another set of resources for building AI tools to benefit radiology.