Predicting Single-Unit Activity from Local Field Potentials with LSTMs.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Biomedical Engineering

Background:

• Single-Unit Activity (SUA) is crucial for understanding neural processing and Brain-Computer Interfaces (BCIs).
• Local Field Potentials (LFPs) offer a more accessible signal but their predictive power for SUA is debated.
• Macaque Primary Motor cortex is a key area for motor control research.

Purpose of the Study:

• To quantify the extent to which Long Short-Term Memory (LSTM) decoders can predict SUA from LFPs in the macaque Primary Motor cortex.
• To evaluate LFPs as a potential proxy for SUA in both fundamental neuroscience research and BCI development.
• To identify factors influencing LFP-based SUA prediction accuracy.

Main Methods:

• Utilized Long Short-Term Memory (LSTM) neural networks to decode SUA from LFP signals.
• Investigated prediction performance across different electrode implant locations and individual animals.
• Analyzed the relationship between LFP power in specific frequency bands (0-10, 10-20, 40-50Hz) and SUA prediction accuracy.
• Compared the efficacy of bipolar versus unipolar LFP recordings.

Main Results:

• Prediction quality varied significantly based on implant location and animal.
• Higher prediction accuracy correlated with increased power in specific LFP frequency bands.
• Bipolar LFPs demonstrated greater informativeness for SUA prediction compared to unipolar LFPs, suggesting effective noise reduction.
• Individual best bipolar LFPs outperformed the combined use of all unipolar LFPs, indicating potential for data compression.

Conclusions:

• LFPs are moderately predictive of SUA in the macaque Primary Motor cortex.
• Bipolar LFP recordings and strategic channel selection show promise for improving LFP-based BCI performance.
• Further research and signal processing advancements may enhance the utility of LFPs as a proxy for SUA.