Decoding movement direction from cortical microelectrode recordings using an LSTM-based neural network.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Brain-machine interfaces (BMIs) enable communication via neural signals.
• Kalman Filter (KF) is the standard for decoding movement from neural data.
• Existing methods face challenges in decoding accuracy and computational efficiency.

Purpose of the Study:

• To implement and evaluate a multi-layer long short-term memory (LSTM)-based artificial neural network (ANN) for decoding neural signals in BMIs.
• To compare the decoding performance of the LSTM model against the traditional KF model.
• To assess the impact of dimensionality reduction on decoding accuracy and computational complexity.

Main Methods:

• Collected motor cortical neural signals from a nonhuman primate (NHP) using a microelectrode array (MEA).
• Trained and applied a multi-layer LSTM-based ANN for decoding neural signals during a directional joystick task.
• Compared LSTM model performance against a standard KF model.
• Investigated the effect of principal component analysis (PCA) for dimensionality reduction.

Main Results:

• The LSTM model achieved significantly higher decoding accuracy (mean correlation coefficient = 0.84, p < 10^-7) than the KF model (0.72).
• Dimensionality reduction using PCA slightly decreased accuracy for both models (LSTM: 0.80, KF: 0.70) but substantially reduced computational load.
• LSTM demonstrated superior performance in decoding joystick trajectories from neural signals.

Conclusions:

• LSTM-based ANNs show significant promise for enhancing decoding accuracy in BMIs.
• The findings suggest a potential improvement for communication and control in paralyzed individuals using advanced AI models.
• Dimensionality reduction offers a trade-off between accuracy and computational efficiency for BMI applications.