Neural Networks for Modeling Neural Spiking in S1 Cortex

Area of Science:

• Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Somatosensation involves touch and proprioception, with less known about proprioceptive cortical signals compared to touch.
• Developing brain-machine interfaces for proprioception lags behind those for touch.

Purpose of the Study:

• To model the relationship between neural firing rates in somatosensory cortex area 2 and arm movement kinematics using Artificial Neural Networks (ANNs).
• To understand how kinematic variables interact to produce proprioceptive responses.
• To explore the potential for a proprioceptive brain-machine interface.

Main Methods:

• Utilized Artificial Neural Networks (ANNs) to model single-neuron firing rates in somatosensory cortex area 2.
• Trained ANNs with various combinations of input (kinematic variables) and output (neural firing rates) to understand variable interactions.
• Investigated the impact of joint angles and muscle lengths on network performance.

Main Results:

• The inclusion of joint angle and/or muscle length data significantly enhanced the prediction accuracy of neural firing rates.
• ANN models suggest that area 2 neuron firing rates are more closely tied to peripheral sensor activity than to hand position.
• Numerical experiments identified sensitivities of ANN models to training design and hyper-parameters.

Conclusions:

• Neural firing rates in somatosensory cortex area 2 may better reflect peripheral sensor activity than extrinsic hand position.
• The study provides a foundation and tools for using machine learning to analyze S1 neural activity.
• Results offer new insights into the representation of limb motion in the brain and inform future proprioceptive interface development.