Cross-subject decoding of eye movement goals from local field potentials.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Cross-subject decoding of brain signals like local field potentials (LFPs) is crucial for brain-computer interfaces (BCIs).
• Adapting models trained on one subject's data to another subject's data presents a significant challenge due to variations in neural representations.

Purpose of the Study:

• To develop and evaluate a novel supervised transfer learning technique, termed data centering, for improving cross-subject decoding of motor intentions from LFP signals.
• To adapt feature spaces between source and destination subjects for more robust BCI applications.

Main Methods:

• Proposed a supervised transfer learning method called data centering to align source and destination feature spaces.
• Utilized linear transfer functions to model the deterministic relationship between feature spaces.
• Developed a data-driven approach for estimating linear transfer functions using class-conditional moments.

Main Results:

• Achieved peak cross-subject decoding performance for eye movement intentions in macaque monkeys.
• Demonstrated substantial performance improvement compared to random choice decoders.
• Showed that data centering outperforms standard sampling methods, particularly with imbalanced training datasets.

Conclusions:

• Data centering is a viable and novel technique for reliable LFP-based cross-subject brain-computer interfacing.
• The proposed method offers a promising approach for developing advanced neural prostheses.