Supervised segmentation of microelectrode recording artifacts using power spectral density.

Area of Science:

• Neuroscience
• Signal Processing

Background:

• Accurate detection of clean signal segments in extracellular microelectrode recordings (MER) is crucial for high signal-to-noise ratio.
• Current methods often rely on unsupervised change-point detection, which can be less accurate.

Purpose of the Study:

• To present a novel supervised method for classifying artifacts in MER data.
• To evaluate the performance of this supervised method against existing unsupervised approaches.

Main Methods:

• A supervised artifact classification method was developed using power spectral density (PSD).
• The method was evaluated on a database of 95 labeled MER signals.

Main Results:

• The proposed supervised method achieved a test-set accuracy of 90%.
• This accuracy is comparable to human annotation accuracy (94%).
• Unsupervised methods achieved approximately 77% accuracy on both training and testing data.

Conclusions:

• Supervised artifact classification based on PSD offers a significant improvement over unsupervised methods for MER data.
• This technique enhances the reliability and quality of MER studies by improving signal segment detection.