Meeting brain-computer interface user performance expectations using a deep neural network decoding framework.

Area of Science:

• Neurotechnology and Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Brain-computer interface (BCI) neurotechnology offers potential to mitigate paralysis-related disability by translating neural activity into assistive device control.
• Key end-user requirements for BCI systems include high accuracy, minimal setup, rapid response, and multifunctionality, largely dictated by the neural decoding algorithm.

Purpose of the Study:

• To introduce a novel deep neural network decoding framework for BCI systems designed for discrete movements.
• To address critical performance characteristics: accuracy, sustained performance, response time, and multifunctionality.

Main Methods:

• Developed and implemented a deep neural network decoding framework for BCI.
• Utilized intracortical data from a participant with tetraplegia for offline analysis.
• Employed unsupervised updating and transfer learning techniques for decoder maintenance and enhanced functionality.

Main Results:

• The deep neural network decoder demonstrated high accuracy and sustained performance for over a year without daily retraining.
• The decoder exhibited faster response times compared to existing methods.
• Real-time application enabled a participant to control functional electrical stimulation (FES) for paralyzed forearm reanimation and object manipulation.

Conclusions:

• Deep neural network decoders represent a significant advancement in BCI technology performance.
• This framework can enhance the clinical translation of BCIs, improving assistive device control for individuals with paralysis.