Proper reference selection and re-referencing to mitigate bias in single pulse electrical stimulation data

Single pulse electrical stimulation experiments produce brain stimulation evoked potentials used to infer brain connectivity. The choice of recording reference for intracranial electrodes remains non-standardized and can significantly impact data interpretation. When the reference electrode is affected by stimulation or evoked brain activity, it can contaminate the brain stimulation evoked potentials recorded at all other electrodes and influence interpretation of findings.

This specific issue is highlighted in intracranial EEG datasets from two subjects recorded at separate institutions. We present several intuitive metrics to detect the presence of reference contamination, based on artificial similarity between all channels. We also offer practical guidance on mitigating contamination, by switching to a more neutral reference electrode, or by post hoc re-referencing, per stimulation site, to an adjusted common average that is optimized for bias and noise.

Either switching the reference electrode or re-referencing to an adjusted common average effectively mitigated the reference contamination issue. This was evidenced by metrics that indicated increased variability in the latencies and response durations of brain stimulation evoked potentials across the brain, and by increased similarity between experimental runs after re-referencing.

Overall, this study demonstrates the necessity of clear quality checks and preprocessing steps to ensure accurate interpretation of single pulse electrical stimulation data, and it provides a set of statistics and tools to achieve this.