Data processing techniques impact quantification of cortico-cortical evoked potentials

Cortico-cortical evoked potentials (CCEPs) are a common tool for probing effective connectivity in intracranial human electrophysiology. As with all human electrophysiology data, CCEP data are highly susceptible to noise. To address noise, filters and re-referencing are often applied to CCEP data, but different processing strategies are used from study to study.

We systematically compare how common average re-referencing and filtering CCEP data impacts quantification.

We show that common average re-referencing and filters, particularly filters that cut out more frequencies, can significantly impact the quantification of CCEP magnitude and morphology. We identify that high cutoff high pass filters (> 0.5 Hz), low cutoff low pass filters (< 200 Hz), and common average re-referencing impact quantification across subjects. However, we also demonstrate that the presence of noise may impact CCEP quantification, and preprocessing is necessary to mitigate this. We show that filtering is more effective than re-referencing or averaging across trials for reducing most common types of noise.

These results suggest that existing CCEP processing methods must be applied with care to maximize noise reduction and minimize changes to the data. We do not test every available processing strategy; rather we demonstrate that processing can influence the results of CCEP studies. We emphasize the importance of reporting all processing methods, particularly re-referencing methods.

We propose a general framework for choosing an appropriate processing pipeline for CCEP data, taking into consideration the noise levels of a specific dataset. We suggest that minimal gentle filtering is preferable.