Strengths & Weaknesses of RANSAC applied to Epidural & Intracortical Recordings

Area of Science:

• Neuroscience
• Signal Processing
• Biomedical Engineering

Background:

• Invasive neural recordings, such as epidural microelectrocorticography (μECoG) and intracortical microelectrode arrays (MEA), are crucial for understanding brain function.
• Data quality is paramount for accurate analysis, but faulty channels can introduce significant noise and artifacts.
• The Random Sample Consensus (RANSAC) algorithm offers a potential solution for automated artifact detection and correction.

Purpose of the Study:

• To evaluate the efficacy of the RANSAC algorithm for detecting and correcting bad channels in invasive neural recordings.
• To assess RANSAC's performance on high-resolution modalities: epidural μECoG and intracortical MEA.
• To provide guidance on integrating RANSAC into preprocessing pipelines for neural data.

Main Methods:

• Applied RANSAC to simulated and real neural data from epidural μECoG and intracortical MEA.
• Tested RANSAC under three scenarios: epidural signals, intracortical signals, and a noise-only benchmark.
• Conducted spatiotemporal analysis to evaluate RANSAC's impact on signal integrity and noise reduction.

Main Results:

• RANSAC effectively identified and interpolated faulty channels in epidural μECoG data with optimized parameters, significantly improving signal quality.
• Intracortical MEA signals showed minimal improvement with default RANSAC parameters due to lower inter-channel correlation.
• Optimizing RANSAC parameters for epidural data on intracortical recordings led to the attenuation of legitimate signal features, highlighting the need for modality-specific tuning.

Conclusions:

• RANSAC can substantially enhance data quality in epidural neural recordings by improving signal-to-noise ratio and inter-channel correlation.
• The algorithm's effectiveness is modality-dependent, requiring careful parameter optimization for different invasive recording types.
• RANSAC presents a valuable tool for offline and potentially real-time preprocessing of invasive neural data, particularly for epilepsy research and closed-loop applications.