EEG Preprocessing Order Can Permanently Distort Brain Networks

The order in which researchers run EEG preprocessing steps can permanently distort the brain networks their analyses report. That is the headline finding of a new study carried by Bioengineer.org, and it lands squarely on a practice most electrophysiology labs treat as routine housekeeping.

Most scalp EEG pipelines share a familiar skeleton: artifact rejection, filtering, re-referencing, interpolation of bad channels, segmentation. What differs between labs, and often between two students in the same lab, is the sequence. The new work indicates that sequence is not a neutral bookkeeping choice. Reorder the steps and the functional connectivity estimates that emerge — the networks themselves — change, and in ways later stages cannot undo.

The word that matters here is "permanently." An early step applied in the wrong position can alter the signal such that downstream corrections operate on already-distorted data. The error propagates. The final network graph then reflects the pipeline as much as the brain.

What does this change for the reading room and the lab?

For quantitative EEG work — source localization, connectivity mapping, network metrics used in epilepsy evaluation, sleep research, and psychiatry biomarker studies — the practical answer is documentation. If two pipelines produce different networks from the same recording, then published connectivity differences between patient groups may partly reflect methodological divergence rather than physiology.

The finding argues for three habits:

• Report the exact order of preprocessing steps in methods sections, not merely the list of steps.
• Keep the pipeline fixed across all groups and sessions within a study.
• Treat cross-study replication failures with pipeline sequence as a first suspect, not a last resort.

This is an analysis-stage result, not a device story. No new hardware, no FDA clearance question, no patient-facing tool. It concerns what happens after the electrodes come off — the part of the workflow where standardization has always been weakest.

Why sequence beats checklist thinking

A checklist implies that completing each step guarantees a clean result. The study's logic is less forgiving. Filtering before artifact rejection, or re-referencing before interpolation, changes what each later step sees. The operations are not commutative. Any analyst who has watched independent component analysis behave differently after a high-pass filter has already met this problem informally; the new work gives it systematic weight.

For journals, the implication is straightforward. Reviewers of connectivity and network papers should ask for the pipeline order the way they now ask for sampling rates and montage. The cost of the question is a sentence in the methods. The cost of not asking is a literature whose network findings cannot be compared across labs.

The study is computational work on EEG data rather than a clinical trial, so it changes practice at the analysis workstation before it changes anything at the bedside. It does, however, set a research agenda: establishing which orderings are safest for which signals, and building preprocessing tools robust to sequence effects. Until those results arrive, the safest assumption is that your pipeline order is part of your result.