We got our inspiration from the fact that about 75% of adults do not know that they have ADHD. As we explored the possibilities with the Muse headband, we started looking into the potential for it with diagnosing ADHD. We researched the different types of waves the Muse detects. We learned that the Beta and Theta waves are the most prevalent for detecting ADHD and that the ratio between beta and theta scores could be used to diagnose ADHD.

What it does

It checks brain patterns while the user is watching a video that requires selective attention. As the user is trying to count the number of times the people wearing white pass the ball, the patterns in their beta and theta wave activity will be tracked using the Muse headphones. For each of the values in the graph, a theta to beta ratio is calculated and using error bounds, a percentage for ADHD is determined. Based on the data collected, it then tells the user how high their probability of having ADHD is and gives the user percentages so that the user knows the uncertainty associated with the result.

How we built it

We used Android Studio to connect the Muse with our app. Muse had an open source API for android and we analyzed the sample code produced as well as the documentation and determined how we were going to collect the theta and beta scores. The reason we wanted theta and beta scores is because the studies on ADHD diagnosis we read stated that the theta to beta ratio is key to diagnosing ADHD because theta waves are associated with daydreaming, drifting, and losing attention whereas beta waves have to do with concentration. With the help of online APIs and documentation, we are able to acquire the data through bluetooth connection. After figuring out how we would detect the brain wave oscillation patterns, we then proceeded to decide what the selective attention quiz should be. We settled on the classic video involving people passing a ball and the user having to count the amount of times the ball is passed. With this, we were able to find how concentrated and easily distracted our user was which was key to diagnosing.

Challenges we ran into

We had some challenges understanding the Muse API, since they were not fully developed. Another challenge we faced was being able to distinguish the different raw data types that the Muse outputted. Oftentimes in spite of requesting Relative Theta Scores we seemed to be getting results that made little to no sense. These came from errors that were very difficult to detect in our code. The headband proved to be extremely difficult to work with due to occasional malfunctioning which came from a lack of charge. Additionally, while the headband collected accurate data for the most part, sometimes it would output zero or NaN.

Accomplishments that we're proud of

We are proud of our ability to collect the beta and theta waves which we were able to use to create TBR values (Theta to Beta Ratio) and estimate the probability of having ADHD. We are also proud of being able to decipher the Muse API and figure out how to collect various types of beta and theta waves which included relative, absolute, and score-based quantities. These different values helped us get a more reasonable and accurate average for the ratios. Finally, we are proud of creating realtime graphs for the values because the API was designed more for actual numbers so we were happy with the graphs we were able to create in order to help the user better visualize their brain wave patterns.

What we learned

We improved our understanding of Java and Android Studio. We became much more familiar with the platform and its potential pitfalls. We learned how to use different APIs that are built for Android Studio and employing external hardware and taking advantage of it in order to do potential ADHD patients a service. We became more familiar with how different waves in the brain interact when various tasks are being performed. We came with a limited neuroscience background so we were able to learn a lot by using these APIs to gather information about the patterns taking place when someone is performing a selective attention task.

What's next for ADDiagnosis

For the future of ADDiagnosis, a bigger dataset could be used to create more reasonable certainties and weights. Because this was a 24 hour hackathon, we had to use error bounds which were collected with observations from small datasets but for future calculations, we would like more accuracy and less percent error. For this hackathon, we didn't expect for extremely accurate results to be produced.

Share this project: