Deep Residual Learning for Classification of X-Rays

Inspiration

Collectively our team is interested in medical applications of programming for the greater good of society.

In underdeveloped countries, medical misdiagnosis and lack of access to medical specialists can be a real problem.

Our goal was to create a website that would aid doctors and patients in interpreting X-rays for the purpose of disease detection in these countries.

What it does

X-ray images uploaded to our website are run through our deep convolutional neural network which outputs the probability that the patient has certain disease.

How we built it

We built our deep learning convolutional neural network using Keras, TensorFlow, and Python.

Challenges we ran into

Because of the specificity of our neural net, improving accuracy of the classification was difficult.

There were no quick fixes that we could google. Any issues we encountered had to be met with improving our understanding of the behavior of the layers of the neural net to then infer the solution.

Accomplishments that we're proud of

We are proud to have been able to complete this project while navigating uncharted territory and learning and applying new skills.

What we learned

It was a challenge, but we were able to persevere on a deadline, by collectively making compromises and managing our team’s workflow and timeline.

What's next for Deep Residual Learning for Classification of X-Rays

We look forward to improving samples, distribution, normalization techniques, and accuracy in medical diagnosis.

Built With

keras
python
tensorflow

Submitted to

VandyHacks IV

Created by

I worked on automating dataset sorting and directory building in Python. Additionally, I used Node.js and Bootstrap to work on front end development, though due to time it did not end up getting implemented.

lucasRemedios
I worked on the front and back end of our app's landing page, using Node.js for the backend and standard HTML and CSS for the client UI. I used node to implement a file uploader for the medical images that would be analyzed by the neural network and displayed the image onto the browser, this being my first time using node and javascript. I also used node to implement keras.js to analyze the x-ray images and return the results in browser. Unfortunately, this was the major roadblock that ate up the remainder of our time. Still, I have a much greater knowledge of Javascript and Node, as well a better appreciation for those who use them.

Zube Molokwu
I worked on the Convolutional Neural Network. I used Keras and Python for the implementation. I also helped with the training script that loaded the data into the networks's model. I learned about handling big data and batching input for use by the network.

Bishoy Boktor
project manager, deep learning engineer, clean/organize data, normalization tech.

Samuel Remedios