Firmware for epileptic seizure detection using ML
This project was created during the BioMedTech Baja Hackathon 2022 related to Medical Devices.
Our Team (MedCreators) proposed a device capable of detecting epileptic seizures, sending an alert and calling to emergency contacts. One of the main highlights of this project is the use of on-device machine learning to do the inference on the device without the need of sending raw data to external devices to process the IMU data.
More info about the project
Info about the hackaton:
We collected data using a board with an ESP32 microcontroller. This is supported by the Edge Impulse Platform and a firmware was adapted for that task.
Two data classes were defined, a normal human behavior (idle, walking, running, dancing ..) and when the seizure occurs. 10 minutes were collected for each class, simulating seizures to test the concept. Extra samples were collected for the test set, having a 90/10 distribution.
We can see how the model can differentiate between the two classes using the test data
idf.py build
-
The inference process uses 2 seconds of IMU data. The frequency of the inference can be defined depending on the user's needs. We played with values between 50 to 200 ms.
-
A method to activate the seizure flag was created considering a continuous number of inferences with a seizure as an output, to avoid false positive peaks.
-
When the seizure flag is active, a window of 4 seconds is opened to be able to deactivate the alert by the user pushing a button in case of a false positive. During this time the device is vibrating intermittently to alert the user.
-
If the button is not pressed, the flag to send the alert is activated and is sended via wifi in our demo. This alert is used by a service to make a call to an emergency contact.
This project was done in a weekend, is just a proof of concept, but can be used as a base for other related projects.
