blog

Now live: a new kind of data challenge


by Isaac Slavitt

Just this week we launched a new $100,000 challenge where competition-tested algorithms are just the beginning. The Concept to Clinic challenge will build on the winning solutions from this year's Data Science Bowl presented by Booz Allen Hamilton and Kaggle to bring machine learning advances to the front lines of lung cancer detection.

The goal of this challenge is to help put deep learning in the hands of practicing radiologists, detecting lung cancer nodules in early screening CT scans and displaying the results in useful ways for clinical use. Lung cancer kills more people each year than any other cancer and early detection makes a drastic difference in 5 year survival rates, so this challenge is tackling a problem area where years of human life are at stake.

The way it works is that we're wrapping a collaborative competition layer around an open source software project. Instead of submitting numeric predictions, contributors will be submitting code patches (pull requests) to our repository, and will get awarded points as those changes are adopted into the master branch. There is a $100,000 prize pool from the Addario Lung Cancer Foundation, and many in-kind prizes from sponsors like Docker, Travis CI, and Amazon AWS.

This project is not just about tuning model parameters. The predictive models and training pipelines are only one part of the challenge prototype, which is scoped to be a minimum viable product for actual clinical testing. As such, the user interface, feature set, and overall product design is actually in scope for contributors to iterate on and improve. The prototype software from this challenge will be tested by participating clinicians, and then participants will take the clinical feedback into account and continue to build out and hone product features until the challenge ends in early 2018.

We're excited to see how contributors will push this project to the finish line, and we plan to write much more about this in over coming weeks and months.

Interested in joining? Head over to the challenge site and sign up today!