OsteoX

Inspiration

Last week, we read an interview of Dr. Sue Desmond, CEO of the Bill & Melinda Gates Foundation, who talked about the use of technology as a tool to improve outcomes in health.

During the opening ceremonies of YHack, a speaker prompted "What's next?" and thus, insights started developing in our minds that how can we make something novel which would increase the effectiveness of healthcare delivery. We decided to focus on data-driven diagnosis, and that's how this project came into being.

What it does?

Before we discuss how our system facilitates diagnosis, let us talk about the challenges faced by traditional medicine in diagnosing osteoporosis and related bone disorders:

There's no way for human eyes to sense these disorders beforehand – until a fracture gets developed.
Once a fracture develops, the probability of complexity in the future increases substantially.
Research suggests that after reaching the age of fifty, 1 in 2 women will have a future fracture related to osteoporosis.

Why Kinect?

Kinect allows us to rapidly develop and deploy a working prototype that can process massive amount of data to give real-time diagnosis. We ourselves wouldn't recommend using this project in its current form for medical practice as a lot of things can be improved.

The purpose of this demo is to attract attention towards increasing importance of computing in medicine and introduce a novel way how computing can aid medical professionals reach better conclusions.

Medical professionals cannot be replaced by computers – but data-driven tools like these would certainly help caregivers develop insights that substantially improve the quality of care they provide, thereby helping ensure that lives are not affected by preventable diseases such as osteoporosis.

Perhaps given the right resources, we can evolve this into a market ready solution for out-of-the-box deployment.

How it works?

On its initial run, the application works in a "learning mode" that allows it to gather vast amount of data when the Kinect sensor is placed strategically. This data is then input into a Machine Learning library that churns out certain parameters which define the normal posture and skeletal configuration of a human being.

On subsequent runs, any obtained data is compared with the initially obtained model in real-time. Cases that vary widely or that can be matched with known parameters of a positive subject are tagged by a large red blip hovering on the area that is statistically anticipated to be affected. Though this leaves a gap to develop false positives, it makes the possibility of a false negative extremely low, which is far better than having no early warnings at all.

Challenges we ran into

The most difficult challenge we faced throughout the journey of developing our project was that none of us were well-versed with Machine Learning and Neural Networks, which we were able to overcome by the use of a machine learning library.

Accomplishments that we're proud of

We feel some sense of pride that our work was able to enhance the accuracy of Kinect's skeletal model generation by using its depth sensing capabilities. We also feel happy that we have built something that has the potential to improve lives.

What's next

Hack
Prototype
????
Revolutionize Medicine.

While it would be too naive to expect that our demo would revolutionise medicine overnight, we see this as a significant starting point in improving the field of medicine. We look forward to guidance from medical professionals, who would help us understand their needs and develop a platform that is accurate enough to be actively deployed across healthcare delivery systems around the world.