Compete in the MICCAI 2022 Federated Learning Breast Density Challenge
Submit your federated learning (FL) algorithm to the Breast Density FL Challenge! Data scientists, informaticists, and medical physicists are invited to develop the best, most generalizable models for breast density estimation using distributed or federated learning.
During the challenge, participants will develop, train, and test models against digital mammographic imaging screening trial (DMIST) data from more than 33 institutions totaling to 100,000 images from more than 21,000 patients. Participants can submit their models for ranking and validation starting on August 5, 2022. Winners will share their proceedings at the 2022 Medical Image Computing and Computer Assisted Intervention (MICCAI) Conference.
Organized by Harvard Medical Schools’ Mass General Brigham, University of Colorado, NVIDIA, the America College of Radiology (ACR), and NCI, this challenge aims to create generalizable models for breast density estimation that can be used across different systems. Distributed and federated learning are popular approaches to making multi-institutional data sets available for analysis without the need for data sharing. However, there is much to be learned about how best to implement these techniques with heterogenous data. Challenge co-organizer Dr. Keyvan Farahani explained that, “Federated learning in medical imaging has gained significant popularity over the past several years, mainly because, in this approach, one handles issues such as patient privacy and data security by keeping the data private. Although our interest is in public data sets and the related developments, it’s important for us to be aware of other approaches that address artificial intelligence in medical imaging without the requirement for data sharing.”
Algorithms that can better estimate breast density can also play an important role in the diagnosis and early detection of breast cancer. Traditional imaging methods are less likely to detect cancer in patients with dense breast tissue. As a result, more intelligent imaging may be needed to accurately assess their risk of cancer. Estimating breast density at the beginning of a treatment plan could help clinicians better balance the benefits and risks associated with additional medical imaging procedures. This effort by ACR/NCI/NVIDIA is one of the first of its kind in healthcare/medical imaging to promote the development and fair evaluation of different FL algorithms.