NCI supports developing technology that will improve breast cancer detection rates and lower associated costs. In the recent NCI-supported Digital Breast Tomosynthesis Cancer Detection Challenge (DBTex), 8 participating teams built artificial intelligence (AI) algorithms with high sensitivity for finding lesions on digital breast tomosynthesis (DBT) images.

The more detailed images from DBT help radiologists find lesions, but the scans take longer to read. The goal of this challenge was to benchmark AI algorithms that aim to read DBT scans as well as radiologists, given the limited data set. The AI could influence lower costs and time for detecting breast cancer.

Challenge participants used deep learning—a type of AI—to train their algorithms. These algorithms:

• gave promising breast lesion detection performance on DBT images.
• will encourage additional research because the predictions and code for selected methods are now publicly available.

While all the results were promising, some submissions performed better than others because of:

• the specific model participants used.
• changes the teams made to their methods.
• the training data used (e.g., some participants used only the publicly available data set, while others used additional data sets).

If you’re interested in accessing the DBTex challenge data set to replicate or build on the findings, it is available in the Imaging Data Commons. This is a repository of the NCI Cancer Research Data Commons, a cloud-based data science infrastructure that connects data sets with analytics tools.

For more about the challenge, methods, and results, read the full JAMA article.