Machine learning algorithm for decoding multiple subthalamic spike trains for speech brain-machine interfaces.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computational Linguistics

Background:

• The subthalamic nucleus (STN) plays a crucial role in motor control and has been implicated in speech production.
• Restoring communication for individuals with severe paralysis, such as those with locked-in syndrome, is a significant challenge.
• Brain-machine interfaces (BMIs) offer a potential solution by translating neural activity into intended actions or speech.

Purpose of the Study:

• To decode speech features from single-neuron electrical activity in the human STN.
• To assess the number of STN neurons required for accurate speech decoding in brain-machine interfaces (BMIs).
• To evaluate decoding performance across speech production, perception, and imagery.

Main Methods:

• Intraoperative single-neuron recordings from the STN of 21 Parkinson's disease patients during deep brain stimulator implantation.
• Utilized machine learning algorithms, including a novel sparse decoder (SpaDe), to analyze speech-related neural firing patterns.
• Patients produced, perceived, or imagined five monophthongal vowel sounds during recordings.

Main Results:

• The sparse decoder (SpaDe) outperformed other algorithms in decoding speech production, perception, and imagery.
• Achieved 100% accuracy in decoding speech production, 96% for perception, and 88% for imagery.
• Demonstrated a linear relationship between decoding accuracy and the number of neurons used, particularly for perception.

Conclusions:

• Single STN neurons encode sufficient information for high-accuracy speech decoding.
• This research represents a significant advancement towards developing speech BMIs for communication restoration.
• Provides crucial insights into the neural basis of speech processing and the feasibility of STN-based BMIs.