Machine learning decoding of single neurons in the thalamus for speech brain-machine interfaces

Area of Science:

• Neuroscience
• Biomedical Engineering

Background:

• The ventralis intermediate nucleus (Vim) of the thalamus plays a role in motor control and may be involved in speech.
• Decoding neural activity is crucial for developing effective brain-machine interfaces (BMIs).

Purpose of the Study:

• To decode single neuron firing patterns in the left Vim related to speech production, perception, and imagery.
• To determine the number of Vim neurons required for accurate speech decoding in BMIs.

Main Methods:

• Intraoperative single neuron recordings from the left Vim in eight neurosurgical patients.
• Utilized the Spade decoder, a machine learning algorithm based on sparse decomposition, to analyze neural firing patterns during speech tasks.
• Recorded neural activity during speech production, perception, and imagery of vowel sounds.

Main Results:

• The Spade decoder achieved high accuracies: 100% for production, 96% for perception, and 92% for imagery (chance level: 20%).
• Decoding accuracy showed a logarithmic relationship with the number of neurons used for all speech aspects.
• Speech production decoding consistently yielded higher accuracies than perception and imagery.

Conclusions:

• Single neuron activity in the left Vim is a promising input source for speech-related BMIs.
• Understanding the neuron count needed for decoding accuracy is vital for planning BMI implantation for patients with speech impairments.