科研 Skill

EEG预处理流程审查

neuro-eeg-preprocessing-review

Designs or audits an EEG or MEG preprocessing pipeline in MNE-Python, EEGLAB, FieldTrip or Brainstorm. Use when choosing or reviewing filter, ICA or epoching settings.

发布者 Codar安全审核通过v0.1.0更新于 2026-11-18

SKILL.md

EEG/MEG preprocessing: settings and order

Preprocessing choices can create or remove effects. Use this order and these defaults unless there is a reason not to, and record every choice.

0. Load the toolbox reference

Read the file for the toolbox in use before writing or reviewing code:

ToolboxReference
MNE-Python (and MNE-BIDS-Pipeline, MEG Maxwell filtering)references/mne.md
EEGLAB (clean_rawdata, ICLabel, ERPLAB)references/eeglab.md
FieldTripreferences/fieldtrip.md
Brainstormreferences/brainstorm.md
Moving between toolboxes, or reviewing a methods sectionreferences/equivalents.md

Each has a working recipe, the toolbox's own traps, and what to report. Loading files of any format: neuro-file-formats.

1. Recommended order

  1. Inspect raw data: units, sampling rate, channel types, montage, events.
  2. Mark bad channels (flat, saturated, noisy; PREP-style criteria) before anything that mixes channels.
  3. High-pass filter (see 2). Optionally low-pass and resample (resampling needs an anti-alias low-pass; MNE applies one).
  4. Line-noise removal: notch at line frequency and harmonics, or spectral methods (ZapLine, CleanLine) when you want to keep the band intact.
  5. Re-reference (average reference only with good head coverage and bad channels excluded or interpolated first).
  6. ICA: fit on a copy high-passed at 1 Hz (more stable), apply the unmixing to the data you analyse (Winkler et al. 2015). Classify components with ICLabel or EOG/ECG correlation; remove only clear ocular, cardiac, muscle or line components.
  7. Interpolate bad channels.
  8. Epoch, then reject or repair epochs (thresholds or Autoreject).
  9. Baseline-correct (ERPs) or not (decoding: baseline correction can move effects; see neuro-bci-decoding-eval).

2. Filter defaults and why

AnalysisHigh-passLow-passNotes
ERP, slow components (P300, N400, CNV)0.01-0.1 Hz30-40 Hzabove 0.1 Hz distorts and can invert slow ERPs (Acunzo et al. 2012; Tanner et al. 2015)
Time-frequency, oscillations0.5-1 Hzabove band of interest
Decodingas low as practical; check time shiftstask dependentzero-phase filters move effects in time
Real-timecausal filters onlysee neuro-realtime-bci-pipeline

Prefer linear-phase FIR (MNE default firwin) with automatic transition bands. Report filter type, order or length, cutoffs, transition bandwidth, and causal vs zero-phase (Widmann et al. 2015). Check the impulse response for ringing if effects sit close to sharp events.

3. Audit checklist for an existing pipeline

  • Units and scaling correct (V vs uV)? Montage matches the cap?
  • High-pass cutoff appropriate for the components analysed?
  • Was a filter applied to epochs instead of continuous data (edge effects)?
  • Reference applied before bad-channel handling? Average reference with few channels?
  • ICA fitted on data with slow drifts, or on too little data (rule of thumb: many times more samples than channels squared)?
  • Components removed on unclear grounds?
  • Rejection thresholds identical across conditions, and trial counts balanced or reported per condition?
  • Baseline window free of stimulus-related activity?

Output

A table: step, setting, reason, reference. Flag each deviation with its likely effect on the user's analysis.

References

Widmann, Schroger & Maess 2015; Acunzo et al. 2012; Tanner et al. 2015; Winkler et al. 2015; Bigdely-Shamlo et al. 2015 (PREP); Pion-Tonachini et al. 2019 (ICLabel); Jas et al. 2017 (Autoreject).

示例提示词

帮我检查这套EEG滤波重参考和ICA的预处理流程设置是否合理。

所属插件

神经科学面向神经科学研究:EEG/MEG、fMRI、BIDS、spike sorting、连接组、脑机接口解码评估、神经刺激安全与公开神经数据检索。

安装该插件会一并获得同领域的 45 个 Skill / MCP。

同插件的其他科研 Skill

常见问题

EEG预处理流程审查能做什么?

Designs or audits an EEG or MEG preprocessing pipeline in MNE-Python, EEGLAB, FieldTrip or Brainstorm. Use when choosing or reviewing filter, ICA or epoching settings。

如何安装EEG预处理流程审查?

在科搭 Codar 左侧点击「插件」,在「组件」页签搜索「EEG预处理流程审查」并安装;在 Claude Code 或 Codex 中可以发送本页的安装提示词,或下载 Skill 包后解压到 ~/.claude/skills 或 ~/.codex/skills。

EEG预处理流程审查需要哪些权限?

该科研 Skill没有声明额外权限;执行具体操作时仍由 Agent 按自身规则请求确认。

EEG预处理流程审查经过官方审核吗?

是。SpeedSkills 收录的每个版本都会记录内容摘要与权限声明,当前版本已通过安全审核。