Stereo-electroencephalography (SEEG) reference based on low-variance signals.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Accurate electrical referencing is essential for interpreting stereo-electroencephalographic (SEEG) recordings.
• Traditional physical references can be noisy, while virtual reference calculations are often time-consuming.
• Developing efficient and reliable SEEG referencing methods is critical for clinical assessment.

Purpose of the Study:

• To evaluate a novel method for calculating SEEG reference signals using low-variance (LV) data subsets.
• To compare the performance of LV-based references against traditional methods, including average and independent component analysis (ICA).
• To assess the computational efficiency and signal quality improvements offered by the LV approach.

Main Methods:

• Analyzed SEEG data from ten patients, focusing on low-variance signal segments.
• Calculated four reference signals: average from white matter (WM) contacts (AVG_WM), average from LV contacts (AVG_LV), ICA from WM (ICA_WM), and ICA from LV signals (ICA_LV).
• Evaluated references by analyzing bipolar signals and average signals from anatomical structures, measuring mutual correlations and outlier correction.

Main Results:

• 91.7% of WM SEEG contacts exhibited below-average variance.
• ICA_LV demonstrated the best performance, while AVG_WM performed the worst.
• AVG_LV provided significant improvements in minimizing inter-structure correlations and correcting outliers with minimal computational cost (0.7870 seconds).

Conclusions:

• Utilizing a low-variance (LV) data subset effectively enhances SEEG reference signal quality.
• AVG_LV offers a fast, straightforward, and stable method for SEEG referencing, outperforming complex ICA methods in terms of speed.
• While ICA_LV shows strong results, its computational demands limit its practical application compared to the AVG_LV approach.