Deep Learning Neural Encoders for Motor Cortex.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Intracortical brain-machine interfaces (BMIs) translate neural activity into control signals for prosthetics and communication devices.
• Clinical viability of BMIs demands highly accurate and robust decoders, but optimization is time-consuming, requiring extensive animal or human experiments.
• Current BMI systems are closed-loop, meaning decoder optimization with offline data cannot capture user adaptation to imperfect decoding.

Purpose of the Study:

• To develop and evaluate deep learning neural encoders for improving closed-loop experimental simulators of BMIs.
• To address limitations of prior neural encoders that failed to capture key features of neural population activity.

Main Methods:

• Utilized deep learning models as neural encoders to synthetically generate neural population activity from kinematics.
• Compared the performance of deep learning neural encoders against prior models in reproducing neural activity features like peri-stimulus time histograms (PSTHs) and population dynamics.
• Assessed the impact of deep learning encoders on neural decoding accuracy using both offline and closed-loop experimental data.

Main Results:

• Deep learning neural encoders significantly outperformed previous models in accurately reproducing PSTHs and neural population dynamics.
• These advanced encoders demonstrated better alignment with neural decoding results from both offline and closed-loop BMI data.
• The findings indicate superior performance of deep learning approaches in modeling neural population activity.

Conclusions:

• Deep learning neural encoders offer a substantial improvement over traditional methods for simulating neural activity in BMIs.
• These enhanced encoders are expected to significantly accelerate the evaluation, optimization, and characterization of BMI decoder algorithms.
• The study paves the way for faster development and clinical translation of more effective brain-machine interfaces.