We claimed a 6th place finish in Kaggle’s machine learning image recognition contest facing off against 661 teams from all over the world!
deepsense.io is very happy to announce that our team was once again at the top of the leaderboard after a major Kaggle competition, sponsored by the California Healthcare Foundation, which was completed on the 27th of July. The main focus was to develop a machine learning technique that will help speed up and automate diabetic retinopathy (DR) screenings.
854 players were working hard for 6 months to come up with a solution to this real-world problem – where diagnosing DR is a time-consuming and manual process that involves a trained physician to examine and evaluate digital color fundus photographs of the retina. Frequently human readers submit their reviews a day or two later, consequently the delayed results contribute to lost follow up, miscommunication, and delayed treatment.
“During this competition we were able to demonstrate how deep learning can be leveraged to revolutionize medicine” says Robert Bogucki, deepsense.io Team Captain. “This was yet another proof of the effectiveness of convolutional neural networks for image recognitionrelated tasks” he adds.
All the teams pushed their automated detection systems to the limits of what is possible producing a whopping 6,999 models with realistic clinical potential. The winning models will be open sourced to maximize the impact such a model can have on improving DR detection.
Read more about the competition here.
About the California HealthCare Foundation
CHCF is leading the way to better care for all Californians, particularly those whose needs are not well served by the status quo. We work to ensure that people have access to the care they need, when they need it, at a price they can afford. CHCF informs policymakers and industry leaders, invests in ideas and innovations, and connects with changemakers to create a more responsive, patient-centered health care system.
Brand Manager at deepsense.io
deepsense.io trademarks at boilerplate