NeuroTechSC has successfully submitted their project into their first NueroTechX competition! NeuroTechSC submitted their project, Boole-pathy, into the open challenge. You can find the submission video here. The following is an overall timeline of each team and their progression throughout the project.

Hardware Team

The hardware team was in charge of making the physical device wearable and to take recordings, providing the raw data for the data team.

In the beginning, various members of the team were unfamiliar with the CAD software (Fusion 360). The use of this software was optimal for the online collaborative environment brought on by the Covid-19 pandemic. After team members familiarized themselves with the CAD software, they began to brainstorm possible designs to enclose the OpenBCI board, its battery pack, and the electrodes that attach to the user’s face. In the early stages of design it was thought that a headset would enclose all the components mentioned above. However, this proved to be too much weight and the final design has the electrodes housed on the headset while the rest of the components are kept on the user’s bicep.

There were several challenges the Hardware team had to overcome that were mainly caused by the pandemic. One difficulty was having multiple iterations of the board enclosure made. This had all the hardware components remain with one team member until the final 3D printed product was made. It had to remain with the person who had access to the 3D printer so corrections to the design could be immediately made. This wouldn't have been an issue regularly but team members were not located in each other's immediate vicinity.

As for the gathering of data, the Cyton board along with the electrodes were interchanged amongst team members who followed these instructions to complete recordings. This was until the components had to spend some time with the 3D printer to run through multiple iterations. Finally, it was handed off to the final team member that demonstrated the headset’s use in the presentation. They put on the finishing touches like attaching copper wire to the headset which was used to stabilize and guide the electrodes with the use of shrinkwrap. This brings us to the final product submitted to the NeuroTechX competition.

Data Team

The data team in NeuroTechSC oversaw the pipeline of the project. They were taking in the data from the hardware component, analyze and filter the data, then sending the data to the machine learning team. Their work was very essential to the outcome of how functional the project turned out to be. The data team first began creating the pipeline from the hardware team using BrainFlow, an application program interface that allows you to acquire data from biosensors, parse it, and then analyze it. It is a powerful tool that allows you to receive data from various boards without having to change your code. The data team decided to write their code in the Python programming language.

The team focused on creating a pipeline that received Electromyography (EMG) data. Because they finished creating the basic pipeline early on in the project, they still had not received any data from the hardware. Therefore, in order to get practice, the team members trained with this model. They began to get practice using different forms of data processing, like normalization, Independent Component Analysis (ICA), and using filters like the Butterworth Filter. Once they began working with actual data, they encountered several problems, the main one being that most of the data was being removed when processed. The team narrowed the problem’s cause to come from the form of data analysis, specifically the IPA they were using. After adjusting and testing, they found that using solely filter, IPA, and standardization yielded better results, removing normalization from their pipeline. With this improved pipeline, the team was able to reach an accuracy of 85 percent.

For the remainder of the project, the data team collected new incoming data from the hardware, each data containing new recordings from different individuals. The pipeline accuracy varied from each recording, so the team tweaked and fixed the pipeline. By the end of the project, the data team was able to reach an accuracy of around 85 percent for all data.

Machine Learning

The machine learning team was in charge of analyzing the data sets and predicting if the data given was a ‘yes’ or ‘no’. In the beginning, the team had to choose between using 1-dimensional or 2-dimensional convolution data. Even though 2-dimensional data shows a lot more, is sometimes less accurate, harder to implement, and less efficient so they chose to use 1D. Next, they split into two teams, one working on the convolutional neural network(CNN) model and the other working on the recurrent neural network(RNN) model. Eventually, the team ended up basing their model on the EEGNet paper which is a CNN for EEG-based brain-computer interfaces. This model yielded better results and after training it for a while, they were able to predict with 85-90% accuracy.

UI

The User Interface (UI) team was in charge of creating a full-stack web interface for the project. The team decided on using Reach, an open-source JavaScript, to create the interface. For the backend development, the team decided to use Flask, which is written in Python. While frontend development is used for web interfaces, backend development is the hidden software that a user does not see but essential to creating a website. For this project, the UI team was required to display a yes/no question, create a countdown, then display detected answer onto the interface.

The UI team began by creating a boilerplate code, which is code that can be reused with little change. This code displayed random questions and an answer, which was used as a placeholder until they received actual data from the machine learning team. As they waited for data, the team members began to familiarize themselves with web interface techniques, like technology stacks which are used to build and run web applications. The team also worked with the data team and its pipeline to create the interface. The team concurrently worked on integrating the frontend interface with the backend as well as creating the 2-second countdown.

After creating their first version of the interface, they were able to successfully connect to the machine learning data and display the prediction onto the web interface. Their next steps were to connect to the hardware and display the port number. The team struggled a bit with this but eventually were able to successfully connect. The UI team then focused on finishing touches to the web interface for the final submission. Latest UI