OurVision

Inspiration

Computer vision is a promising technology for preventative care. However, the current model of deploying computer vision for preventative and early detection creates barriers to access for america's marginalized communities. There are approximately 80 million people enrolled in Medicaid in the US, these are individuals who have limited access to the high tech imaging equipment that current computer vision workflows rely on. These disparities exists in light of the fact that low socioeconomic status communities have been shown to benefit most from preventative care and early detection.

This disparity is not only unjust, it is financially unwise. Two-thirds of all Medicaid beneficiaries receive their care in comprehensive risk-based managed care organizations (MCO). Providing this patient population with accessible computer vision technology can save billions in healthcare costs for MCOs and the government. Our goal is to make computer vision diagnoses available through mobile devices with OurVision, a computer vision diagnostic mobile application platform.

What it does

OurVision is a mobile application that captures clinically relevant images from mobile video files, classifies diseases identifiable by the captured image in real time, provides patient navigation recommendations and engages the patient in capturing post-diagnosis response outcomes.

How we built it

The mobile application feeds data to the deep learning models that we have developed through PyTorch to identify clinically relevant images from video files and classify those images for certain disease profiles.

Challenges we ran into

It was difficult to identify who the payer would be for our service and how we could deploy it onto the field. For the image capture challenge through video, we had to obtain our own training data which meant we had to capture video of our own eyes and that there are too many negative samples in our training data.

Accomplishments that we're proud of

Our model gets high recall and precision scores, which means that we can classify the pictures into useful and useless ones accurately.

What we learned

How to build a CNN to classify pictures.

What's next for OurVision

We have developed a business model canvas and look to pilot our idea with a managed care organization in the coming months. We will focus on product-market fit to see what our customer segment is most interested in, while maintaining our principle goal to serve marginalized communities