S4SDet

Inspiration

We were inspired by the applications of AI in healthcare. Since we have previous experience with implementing AI in healthcare, we wanted to see how else AI can be applied to different healthcare areas.

What it does

Our system measures the Cobb angle from radiographs of scoliosis patients. The input is first fed through a neural network (which computes landmarks on the spine) and then the Cobb angle is calculated from the landmarks output from the neural network.

How we built it

We built it using deep learning. We used Python, PyTorch and Microsoft Azure in our tech stack.

Challenges we ran into

We had some, but not a lot of experience with PyTorch prior to starting this hackaton. During the hackaton, we had to figure out a lot of PyTorch related things. However, the biggest challenge was the small dataset we were dealing with. We had 481 images (and their corresponding landmarks) intended for training. Dealing with a small dataset is always a challenging task, particularly with deep learning.

Accomplishments that we're proud of

We were able to implement a fully functional system from the ground up. We were faced with frameworks we weren't exposed to much before and with plethora of other challenges. We overcame them all and we're proud of that.