Empirical multi-scale thresholding for evoked neural activity denoising

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Evoked potentials (EPs) are crucial biomarkers for neural function assessment.
• Extracting EPs is challenging due to low signal amplitude and background noise.
• Overlapping frequency ranges complicate reliable EP identification.

Purpose of the Study:

• To present a novel framework for estimating noise distribution around EPs without prior assumptions.
• To enable reliable separation of meaningful EP components from background activity.
• To enhance EP detection reliability and accuracy.

Main Methods:

• A multi-scale bootstrap approach statistically characterizes noise and uncertainty.
• The method empirically estimates variability to characterize noise around the mean.
• Application across multiple frequency bands captures dynamic neural variations.

Main Results:

• The method demonstrated improved signal-to-noise ratio (SNR) with lower mean square error (MSE).
• Enhanced visualization and more accurate morphology recovery of real EPs were observed.
• Reduced false detections and preserved EP integrity were achieved.

Conclusions:

• The approach offers improved EP detection and visualization for clinical and research applications.
• It is particularly beneficial for time-limited recordings or low patient tolerance.
• Supports broader applications in neuroscience and neuro-engineering.