Seizure Catcher

Inspiration

Every day, hundreds of thousands of people live in fear of their loved ones suffering a seizure. With the invention of Seizure Catcher (tm) those fears can be put to rest. Keep track of your loved ones from anywhere around the world, and see their live accelerometer readings to reassure yourself that they are not in any trouble.

What it does

Our incredible invention features a heartbeat sensor and accelerometer, to be worn by the at risk individual. If rapid changes are detected in the state of the individual, accelerometer readings are recorded and uploaded to the server, where they are processed by a neural network in order to classify the readings as "Normal" or "Seizure."

Additionally, we have some cool features such as graphs of the readings that are publicly accessible so that anybody in the world can track your movements.

How we built it

The sensor system is fairly straightforward -- an accelerometer and pulse sensor linked to a dev board, which is mounted on the user's arm. Using the onboard wifi module the dev board sends http requests to the server, which has a neural network that was trained on a set of data comprised on people doing regular activities and having seizures.

The neural network searches for the most recent group of results -- approximately 200 readings -- and returns a result to the dev board after the server saves the posted data values into a MongoDB cluster.

Finally, there are a few additional routes that display a dashboard for the user to interact with, complete with graphs of the accelerometer readings in all three dimensions.

Challenges we ran into

The first challenge we encountered was finding a proper dataset -- most medical and research based seizure measurements focused on EEG, ECG or EMG -- we specifically needed movement data because we don't have the equipment to measure anything else.

The second, more major challenge we faced was the ESP32 dev board not connecting to any of the accelerometers that we had available to us. This was consistent across the multiple ESP32's and sensors we tried, and we were only able to fix this by swapping to an arduino and porting our code to match the change in hardware.

Accomplishments that we're proud of

Our graphs are very friendly to anyone whose hearing is impaired. Tabbing into them lets you hold the arrow key and listen to someone's accelerometer readings! You can hear them having a seizure.

What we learned

ESP32's can be very annoying sometimes. We also learned that flask serving html directly is not a great idea and terrible once you try to add js in. Also finding datasets is much harder than we expected, especially when the data sets seem like they might have what we're looking for but ultimately would require a significant amount of operations to derive the data we need.