FullNode Health

FullNode Health detects an incoming seizure with inputted EEG data using a CNN and recommends music that potentially mitigates the symptoms of an impending seizure.

Comment

Inspiration

Epilepsy is the most common neurological disorder, affecting 1% of the U.S. population – over 3 million people. Among them, 20-30% suffer from refractory (drug-resistant) epilepsy, which requires more innovative and creative approaches to management. Our project aims to address this gap by developing a solution that not only predicts seizures but also helps mitigate their impact through music therapy.

What it does

FullNode Health project utilizes machine learning to predict the onset of a seizure based on EEG data. Using a Convolutional Neural Network (CNN), the system processes the EEG data, outputs a probability of whether a seizure is imminent, and recommends a music genre to help reduce stress and calm the patient.

  1. The model classifies the input EEG data into two categories: 1 (seizure) or 0 (no seizure).
  2. If a seizure is predicted with high confidence, the system triggers the music recommendation.
  3. Once the music genre is selected, it's fetched from Spotify in real-time and played for the patient, offering immediate relief.

How we built it

We developed a full-stack solution with a React frontend and integrated machine learning models.

  • The frontend allows users to upload EEG data (in CSV format) for seizure prediction.
  • The backend processes this data through our CNN model to generate seizure predictions and send back the corresponding music recommendations.
  • We used the Spotify API to stream the recommended music directly to the patient’s device in real-time.

Challenges we ran into

Data cleaning and preprocessing proved to be a significant challenge. The EEG data was irregular and required extensive cleaning, as well as rigorous preprocessing steps before it could be used effectively for model training. Learning how to collect the necessary tokens and IDs to connect the app with the Spotify API was time-consuming and challenging. It required an understanding of the authentication process and how to dynamically fetch the right music tracks.

Accomplishments that we're proud of

  • Successfully integrated EEG analysis with real-time music recommendations based on seizure predictions.
  • Developed a working full-stack application that includes the frontend, machine learning backend, and Spotify integration, all of which interact seamlessly.
  • Achieved an initial 65% model accuracy in less than 24 hours, which is a solid start for a complex task.

What we learned

  • Machine Learning Team: We gained hands-on experience in data cleaning, model training, and fine-tuning predictive models. Additionally, we honed our social and collaboration skills while troubleshooting the complexities of working with real-world data.
  • Fullstack Team: We enhanced our skills in full-stack development and learned how to integrate an external API (Spotify) with both the frontend and backend of an application. This allowed us to create a seamless user experience.

What's next for FullNode Health

Our next step is to refine the model and scale it to handle other neurological disorders, where similar approaches could be used to predict episodes and provide tailored music therapy. The ultimate goal is to deploy FullNode Health in hospitals and clinics, making it a useful tool for patients with epilepsy and other disorders that can benefit from therapeutic music interventions.

Built With

Share this project:

Updates