[Eeglablist] Seeking feedback on transparent, researcher-controlled EEG preprocessing QC

[1435893987]

Dear EEGLAB community,
I am developing an EEGLAB-based tool to address a practical gap I repeatedly encounter in EEG preprocessing: the gap between an algorithm flagging a potential problem and a researcher making a defensible final decision.
Many existing methods can identify possible bad channels, artifacts, or low-quality segments. In practice, however, a large amount of time is still spent answering questions such as:
Is this channel or segment truly problematic, or just unusual?
Which evidence should be considered when different QC indicators disagree?
What changed after an intervention such as channel rejection, interpolation, ICA component removal, rereferencing, or segment rejection?
Can the reasoning behind these decisions be reviewed, reproduced, and compared across participants later?
My goal is therefore not to build a black-box pipeline that automatically modifies EEG data. Instead, I am exploring a researcher-controlled QC workspace that can present multiple forms of evidence, preserve source-time and processing history, support before/after inspection, and record the final human decision.
The current prototype focuses on continuous resting-state EEG and includes channel-by-time-window QC evidence, spectral and spatial views, non-destructive supporting evidence, checkpointed processing actions, post-action inspection, and structured exports. I am currently testing these ideas on real EEG data.
I would be especially grateful for your perspectives on the following questions:
In your own workflow, where is the greatest burden after an automated method has flagged a possible problem: interpreting the evidence, deciding what action to take, checking the consequences, documenting the decision, or reviewing many datasets?
When a tool flags a bad channel, artifact segment, or ICA component, what evidence would you need to see before trusting or rejecting that suggestion?
Which preprocessing or QC actions should always remain explicitly researcher-confirmed? Are there actions you would never want a tool to perform silently?
What information would make a preprocessing/QC decision auditable and reproducible for your lab, collaborators, reviewers, or future self?
Are there missing EEGLAB features that would make it easier to inspect, compare, and document preprocessing decisions across datasets?
I would greatly appreciate any practical examples, frustrations, or suggestions. If this problem resonates with others, I would also be happy to invite interested researchers to provide feedback on an early research prototype once its scope is sufficiently stable.
Best regards,
Junming Wang （Henan medical university）
1435893987
1435893987 at qq.com