OsteoCNN

Inspiration

We have seen multiple circumstances of ill and unfortunate people with osteoporosis who are being inhibited in daily life activities by their conditions, to which we wanted to examine and identify the causes and differences of regular people and those who have osteoporosis and what we will build is a model that can accurately predict if someone has osteoporosis or not just by an image of their bones.

What it does

The CNN (convolutional neural network) simplifies the images from layers and self-learns what's accurate or not and then will give an output. It will look at a data set of a bunch of bone pictures of the knee in low, medium, and high levels of bone density loss and will gain pattern recognition to properly predict a person's bone-density loss % based on their knee x-rays. It starts from a standard network that has multiple interconnected layers that receives an input and relays an output to the next layer. Inside each sub-layers, there are filters that are responsible for pattern recognition. These filters are usually a 3-3 block that can detect specified patterns that it's told. The filter will start by its size and move across pixels and try to recognize any specific pattern until all of the available pixels have been analyzed. When multiple filters with multiple patterns are pooled (process of pooling), this creates a layer of pixels for pattern recognition. As more layers of filters are added, they'll be able to start recognizing simple shapes like a square or a window and eventually big infrastructure like houses or buildings.

How we built it

We converted images from the knee x-ray database into np arrays with rgb values. These arrays were classified as normal, osteopenic, or osteoporotic, and then split into training and testing data (300 training samples and 74 testing samples). Our CNN classifier was built using tensorflow keras and then fed the training data and tested on the testing data, achieving 84% accuracy.

Challenges we ran into

Our first and most difficult challenge was to find and come up with a good and realistic idea. We first thought of trying to math out equations that could predict the orbits of a 3 body system, or more commonly known as the n-body problem, which we realized was close to impossible to do. We then thought of comparing habitable planets but we also realized that it wasn't feasible as we cannot compare exoplanets to earth due to each star system having a different star that inputs different characteristics onto its planets, such as the sun with the earth. We then settled on comparing osteoporosis of people on earth and people in space. The second biggest challenge was finding the correct data set with images that we can use to compare as it took all day yesterday for us to try to find a dataset and we weren't successful until the morning of the second day. Overall, the biggest issue for us was to find a reasonable and and accomplishable project on a subject that we were passionate about in a reasonable amount of time.

Accomplishments that we're proud of

We are proud of our perseverance and tenacity to keep on searching and probing for a reliable data set that we can use. Our coordination was a little bit messy at the start but after we found a project we unanimously agreed on, our teamwork and communication became the best we could've asked for and we were able to make significant progress in a short time, which really speaks volume into the trust and understanding that we have been able to instill into each other. We are proud of our results and the model that we've created through our coding and use of convolutional neural network.

What we learned

We have learned that we must be more concise and precise in our planning and decision making so that we reduce the amount of time wasted as well as increase efficiency and opportunities to improve our work and point out/fix mistakes later on. We also learned a lot about the affects of microgravity on the human body and bone strength. More specifically, our bones have osteoblasts that can form new bones and replace old bone tissues when your skeletal system is put under a lot of stress, generally gravitational stress. Our bones also have osteoclasts that dissolves and break down old bone cells and tissues when not placed under stress, which when you're in space without any gravitational pull, your body won't be able to produce osteoblasts and your bones will break down and decay over time. If you were to just lay down and sleep for a whole year on earth, you'll have the same effect of osteoporosis as you were in space since the human body functions on the same basic principle.

What's next for Spatial Osteoporosis

We hope that we will be able to predict results with an accuracy of greater than 60% to be somewhat consistent and reputable. We someday want to allow this idea to be an affordable and accessible construction to everyone who wants to have a health check and prepare in advance and prevent injuries if they learn that they have osteoporosis.