AI image challenge

The growth of artificial intelligence (AI) is expected to provide valuable tools for the field of radiology. That's why we created the AI Image Challenge, a program dedicated to fostering and demonstrating AI and machine learning capabilities.

In these challenges, we encourage artificial intelligence researchers to create applications that perform a defined task according to specified performance measures. The ultimate goal for the program is to demonstrate how machine learning and AI can benefit radiology and improve patient care.

If you have any questions about the program, please email informatics@rsna.org.

Challenges

The AI Image Challenge will be an ongoing competition. Check here for all the latest information and challenge details.

2018: Pneumonia detection

For the 2018 challenge, we invited teams of data scientists and radiologists to use chest x-ray datasets made public by the National Institutes of Health (NIH) to develop algorithms to identify and localize pneumonia.

Prizes

Kaggle, Inc., in partnership with RSNA, will award $30,000 in prizes to the winning entries of the challenge.

How it works

The challenge has two phases and two corresponding datasets: training and evaluation. The training phase ran through October 24. During this phase, participants used the training dataset to develop algorithms that replicated the annotations provided by the radiology observers.

The evaluation phase, from October 25 - 31, is when participants run their algorithms on the testing portion of the dataset. This dataset had no visible annotations.

All results will be compared to the annotations on the testing datasets provided by expert radiology observers to determine the winners. Challenge results will be announced in early November and top submissions will be recognized at the Machine Learning Showcase at RSNA 2018.

For detailed information about this challenge, visit the Kaggle site.

Past challenges

2017: Bone-age assessment

The first AI Image Challenge focused on assessing bone age from pediatric hand radiographs.

Over 250 participants, comprised of radiologists, technology companies, computer scientists, engineers and other medical specialists, competed in the challenge. Participants worked in 29 teams to submit the outcomes of their algorithms. Teams with the most accurate predictions were announced at RSNA 2017.

The winning algorithm provides radiologists with an automated method for analyzing bone age — a significant improvement on current methods which can be cumbersome and time-consuming.