Skip to content

CLAUDE.md

Latest commit

 

History

History
62 lines (54 loc) · 3.15 KB

CLAUDE.md

File metadata and controls

62 lines (54 loc) · 3.15 KB
Raw

CLAUDE.md — BIDS EEG Study Template

Project Overview

BIDS-compliant EEG analysis pipeline template from the NeuroCognitive Imaging Lab (NCIL), Dalhousie University. All study-specific settings live in config.json.

Directory Structure

BIDS/
├── config.json               # Central config — edit this for each study
├── sourcedata/               # Original raw data (write-protected, not shared publicly)
├── sub-???/                  # BIDS-compliant raw datasets per subject
├── derivatives/              # Processed outputs from each pipeline stage
├── stimuli/                  # Stimulus files
└── code/
    ├── 0_init/               # Initialize BIDS structure from config.json
    ├── 1_import_data/        # Convert raw EEG → BIDS format
    ├── 2_preprocessing_erp/  # Filter, ICA, epoch → derivatives/
    ├── 3_visualization_erp/  # ERP plots
    ├── 4_measurement_erp/    # Peak amplitude/latency extraction
    ├── 5_stats_erp/          # Linear mixed-effects models
    ├── 6_classification/     # SVM classification
    └── 7_stats_classification/

Key Configuration (config.json)

  • EEG hardware: Brain Products actiChamp, EasyCap M1 (easycap-M1 montage)
  • Channels: 32 EEG + 2 EOG (HEOG, VEOG); ground at FCz; 32 non-cortical channels dropped during preprocessing
  • Raw format: XDF (.xdf) → converted to BrainVision format for BIDS
  • Preprocessing filter: 0.1–30 Hz (analysis), 0.5–30 Hz (ICA)
  • ICA: Picard method, 25 components, z-threshold 2, random_state 42
  • Epochs: −1 to +1s, no baseline at preprocessing; rereferenced to average
  • n_jobs: 8 (parallel processing)

ERP Components

The temaplte project shows how to measure two common ERP components, the N170 and N400. If you are interested in other components, you can change the settings in config.json to reflect the names, ROIs, and time windows relevant to your study.

Component Peak Lat Window ROI Measure
N170 170ms ±25ms vertex (Cz, CP1, CP2, Pz) neg peak
N400 300ms ±100ms vertex (Cz, CP1, CP2, Pz) mean amplitude
  • Analysis window: −0.1 to 1.0s; baseline: −0.1 to 0s
  • Outlier threshold: 2 SD; p-threshold: 0.05

Tech Stack

  • Python, Jupyter notebooks
  • MNE, mne-bids, mne-bids-pipeline
  • AutoReject (bad segment detection)
  • Picard ICA
  • SVM classifier (scikit-learn): 80/20 split, 10-fold CV

Workflow

  1. Edit config.json with study details
  2. Run code/0_init/0_init_BIDS_study.ipynb to create BIDS structure
  3. Copy raw data to sourcedata/, then run code/1_import_data/ notebooks
  4. Run code/2_preprocessing_erp/1_preprocessing_batch.ipynb (batch over all subjects)
  5. Visualize → measure → stats → classification

Notes for Claude

  • Notebooks iterate over subjects found in the BIDS root directory
  • Derivatives are saved per-stage under derivatives/<stage_name>/
  • config.json fields propagate through all scripts — changes there affect everything downstream
  • The template has EDIT_ME placeholders in config.json that should be replaced for real studies