We have always been interested in the impact of technology on medicine. We came across this paper called DeepSurv link which details the different survival methods used by medical practitioners to understand the effectiveness of various treatment options and thought that it would be awesome to implement these algorithms to calculate the current risk and of and recommend treatment paths to patients having heart/brain disease, breast cancer and diabetes.

What it does

We built a web based interface where the doctor can select one of the four problems and enter in medical data for a patient. We then trained a deep neural network using previous patient data as the training data based on cox proportional hazards to accurately determine the current risk to the patient and the suitable plan for treatment.

How we built it

We built it using a 45 layer Convolutional Neural Network, various HTML templates and the emoji CSS library to make sure we adhered to the 8 bit theme. We used the Amazon Web Services Cloud9 editor to write and test the website and Jupyter Notebooks for writing and training the Neural Network.

Challenges we ran into

We ran into a lot of problem with training the model, as we had limited time and resources to make sure the predictor worked correctly. Also, none of us are seasoned web developers and we spent a lot of time trying to understand material design.

Accomplishments that we're proud of

Implementing a complex machine learning system from a published research paper and presenting in an easy and fun usable manner through a website.

What we learned

How technology can revolutionise the way we deal with various medical challenges in life. We also learnt how to appropriately manage our time and delegate tasks to ensure we completed this large scale project.

What's next for Healthify

We plan to test more models for risk calculation, possibly adding a dashboard for the doctors to evaluate the conditions of their patients in detail. We also plan to expand our system to more common health problems.

Built With

Share this project: