recov.ai

Revolutionizing Musculoskeletal Health

Comment

Inspiration

I (Sejin) personally suffer from a musculoskeletal disorder. I have bulging discs along my lumbar spine which leads to chronic back pain and occasional radiating tingling down my legs. Due to my chronic condition, I regularly visit my physiotherapist for treatment and receive exercise prescriptions. These exercises are for me to do on my own time so I can actually recover my senses and normal range of motion. However, I often forget to do them and when I do them, I'm not sure I'm actually doing them properly. That's why we developed Recov.ai!

What it does

Recov.ai is a physiotherapy exercise monitoring app that keeps track of a patient's progress of recovery over time and nudges them if they are exercises wrong. It counts the reps for a patient's excercise through the webcams and updates the doctor's database each day on the progress. It also reminds the patient to adhere to their regular exercise schedules as it is an important factor for recovery. An additional application for recov.ai is to help patients better measure their blood pressure correctly and in the correct position: assists the patient until they are able to learn the correct position.

How we built it

It uses a deep learning module built on top of TensorFlow, which uses convolutional neural networks to recognize key body parts from the users' webcam feed or mobile app. It then uses the javascript to operate on the data and makes it in a presentable form: a chart and a loggable database that doctors can have a look into, for all their patients.

Challenges we ran into

Going from the ML model prototypes on a python / Jupyter Notebook environment to working with Tensorflow.js and integrating it into a coherent product. We also struggled with charting the data asynchronously at a live rate.

Accomplishments that we're proud of

I am proud of identifying a critical problem in healthcare and building a functional MVP to address it within less than a weekend. By increasing patient adherence, we can improve health outcomes at both the patient and population level.

What I learned

The simplicity of a problem that no one is solving. 10 Starbucks Doubleshots in 24 hours doesn't kill me (it might kill you though...)

What's next for recov.ai

Implementing the exercise recommendation engine based on similar patient profiles.

Share this project:

Updates