A stack LSTM structure for decoding continuous force from local field potential signal of primary motor cortex (M1).

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain Computer Interfaces (BCIs) translate neural activity into control signals for external devices.
• Accurate decoding of force amplitude is crucial for fine motor control in BCIs, such as grasping.
• Continuous motor BCIs enable users to control robotic arms or limbs.

Purpose of the Study:

• To propose and evaluate a stack Long Short-Term Memory (LSTM) neural network for predicting applied force amplitude from Local Field Potential (LFP) signals.
• To compare the performance of the proposed LSTM network against the Partial Least Square (PLS) method for force decoding.

Main Methods:

• Utilized a stack Long Short-Term Memory (LSTM) neural network architecture.
• Applied the LSTM network to Local Field Potential (LFP) signals recorded from freely moving rats.
• Compared LSTM performance with the Partial Least Square (PLS) method.

Main Results:

• The LSTM network achieved a higher average coefficient of correlation (0.73) compared to PLS (0.67).
• The coefficient of determination was higher for the LSTM network (0.54) than for PLS (0.45).
• The LSTM network accurately predicted force values, including zero-force instances, without explicit time lags in input features.

Conclusions:

• The proposed stack LSTM structure accurately predicts applied force from LFP signals.
• The LSTM-based approach offers higher accuracy and potentially faster, more precise BCI systems by avoiding explicit time lags.
• This method advances the development of sophisticated Brain Computer Interfaces for fine motor control.