The Muse band is an amazing gadget. Having a personal, portable and durable EEG reader at the palm of your hands is an exciting idea for hackers. On the other hand, with a little knowledge about neurology, one can build amazing thing useful for humanity. Our goal was to develop an application capable of helping the core problem of some types of epilepsy, which is the absence of physical spasms. They are called absence seizures.

What it does

The Muse Band reads the EEG data from the user. The signal is analyzed and then the state of the user is classified in NORMAL or SEIZURE. The data is saved on a webserver for future checking from the doctor.

How we built it

The system is built using the Muse Band Android API. The application reads and processes the signal. The signal was first analyzed in MATLAB, trying to find the best feature vector. Then, we used R to find a support vector machine able to classify the state, that was later on codified in Java.

Challenges we ran into

We ran into a lot of challenges. The first one was trying to analyze the signal in order to detect the differences between the healthy and the epileptic ones. Also, the lack of epileptic data for our machine learning process was a handicap. We trained the support vector machine using a dataset coming from hospitals, trying to use the signals taken from the same spots that the Muse measures.

Accomplishments that we're proud of

Our biggest accomplishment is the algorithm for seizure detection. Using only 3 features we obtained a classificator with a 83% of accuracy. Also, the implementation of the SVM in Java was really inspiring (you really see that SVM is only a sign test).

What we learned

The usage of machine learning models outside of statistical packages opens a big road ahead for us, since now we know how to interpret some models like LDA, SVM's and LR in other languages.

What's next for NeverAbsent

The future is always uncertain. We hope to develop a better detection algorithm and maybe test the application with real patients. Also, it would be awesome to offer support to different kind of commercial and profesional headbands.

Share this project: