Abstracts

Rationale: A boom in high-frequency EEG and local field potential (LFP) sampling technologies has greatly expanded the volume of data available for neurological analysis in both clinical and research settings. Despite these advancements, the manual identification of high-quality and low-quality, noisy recordings by neurophysiologists is becoming increasingly cumbersome and error-prone, particularly with large datasets featuring unfamiliar sampling rates. We introduce the Multi-Metric Analysis Number (M-MAN), a novel metric designed to automate the identification and selection of high-quality electrode signals within EEG recordings.

Methods: In this study, 64 electrodes, organized into 16 tetrodes, were stereotactically placed in specific brain structures of six rats. EEG data were recorded from these sites at a sampling rate of 32,000 Hz, producing four continuously sampled channels from closely spaced electrodes within each tetrode, which typically exhibited similar signal profiles. These signals underwent prescreening, detrending, and analysis using various metrics related to signal quality and noise.

The core analysis utilized Fast Fourier Transform (FFT) techniques to calculate signal-to-noise ratios (SNR) and assess temporal differences to evaluate signal integrity. These metrics were then equally weighted to compute the M-MAN for each channel.  

The analysis was incorporated into a specialized server stack, enabling automated storage and analysis of the cleaned EEG data. The results were compared across six randomly selected experiments involving four subjects and were independently reviewed as optimal or sub-optimal by three EEG experts to validate the findings.

Results: When validated against expert EEG quality assessments, the M-MAN demonstrated an ability to select the optimal electrode in a tetrode grouping across various test conditions. Notably, the M-MAN achieved effective stratification of “optimal” and “sub-optimal” channels, significantly improving the reliability and efficiency of channel selection.  

The algorithm achieved an accuracy of 97% in correctly identifying the channel that was rated as optimal by at least two-thirds of the reviewers. In comparison, 82% of the channels received a favorable rating from two-thirds of the reviewers. When comparing the algorithm's accuracy to what would be expected by random guessing based on this distribution of ratings, a t-test resulted in a t-statistic of 16.06 with a p-value of less than 0.0001, confirming the significance of the algorithm's performance at the 99% confidence level.

Conclusions: The M-MAN represents a significant advancement in EEG/LFP data processing, offering a powerful tool for the accurate and timely evaluation of electrode signal quality. The integration of M-MAN’s predictions into an automated processing pipeline optimized EEG interpretation by consistently selecting channels with the highest quality signals available. Future research will focus on expanding the automation capabilities of the M-MAN, further reducing the analytical burden on clinicians and researchers and paving the way for more rapid and precise neurological assessments

Funding: This research received no grant from any funding agency.