Data-efficient Causal Decoding of Spiking Neural Activity using Weighted Voting.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain-Computer Interface (BCI) systems offer potential for individuals with neural injuries and hands-free device control.
• Accurate online intention interpretation remains a significant challenge for BCI systems.
• Robust and data-efficient decoding is crucial for real-world BCI applications, despite dynamic neural activity and causality requirements.

Purpose of the Study:

• To develop and validate a causal, data-efficient neural decoding pipeline for predicting user intention.
• To enhance the accuracy and robustness of BCI systems for real-world applications.

Main Methods:

• A novel causal, data-efficient neural decoding pipeline was developed.
• The pipeline classifies neural recordings in short sliding windows.
• It employs weighted voting on predictions for refined, real-time intention decoding.

Main Results:

• The pipeline demonstrated utility in classifying spiking neural activity from the human posterior parietal cortex during a motor task.
• Achieved higher classification accuracy compared to state-of-the-art causal methods.
• Showed robustness to hyper-parameter selection.

Conclusions:

• The proposed causal, data-efficient decoding pipeline significantly improves intention prediction accuracy in BCI systems.
• This approach addresses key challenges limiting widespread BCI adoption.
• The method is effective for real-time neural decoding and robust across different settings.