Deep learning decoding for non-invasive brain-computer interfaces

Brain-computer interfaces (BCIs) have the potential to make life easier for people with movement and speech disorders, especially by allowing them to operate prosthetic limbs and use computers. Additionally, he allowed healthy and disabled people alike to enjoy BCI-based games.

Non-invasive BCIs, which work by analyzing brain waves recorded by electroencephalography, are currently limited by inconsistent performance.

Bin He and colleagues used a deep learning decoder to improve the performance of a BCI that responds to users engaged in the task of tracking objects in two-dimensional space using a cursor.The work will be published in a magazine PNAS Nexus.

Twenty-eight adult participants imagined moving their right hand to move the cursor to the right, moving their left hand to move the cursor to the left, moving both hands at the same time to move it up, and moving both hands to move it down. were instructed to do so, allowing continuous and sustained movement. of virtual objects.

The authors evaluated two different deep learning architectures and a traditional decoder over seven BCI sessions. Both deep learning decoders improved throughout the study and outperformed the traditional decoder by the final session.

With the help of a deep learning-based decoder, a human participant controls a fast, continuously moving computer cursor using an AI-powered non-invasive BCI based solely on brain waves in sensor space. We were able to track randomly moving objects with high accuracy. According to the authors, performance levels can be improved without moving a muscle, which could be a step toward neurally assisted robotics.