Understanding AutoClean Results#
AutoClean creates a structured output directory for each processing run. This guide explains the organization and contents of your results.
Output Directory Structure#
AutoClean creates a folder named after the task used for processing:
output/
└── RestingEyesOpen/ # Named after the task
├── bids/ # BIDS-compliant processed data
├── flagged/ # Data flagged for quality issues
├── metadata/ # Processing metadata and logs
├── logs/ # Detailed processing logs
└── stage_files/ # Exported intermediate data
Directory Contents#
bids/ - Primary Output Data
Contains the main processed EEG data in BIDS format: - Cleaned continuous data - Epoched data (if applicable) - Channel information and metadata - BIDS-compliant file naming and structure
flagged/ - Quality Control Flagged Data
Contains data that triggered quality control flags: - Files with less than 50% epochs remaining after artifact rejection - Data flagged for excessive channel rejection - Processing runs that failed quality thresholds
metadata/ - Processing Information
Contains processing metadata and summary information: - Run summaries and statistics - Processing parameters used - Quality metrics and reports
logs/ - Processing Logs
Detailed logs of the processing pipeline: - Step-by-step processing information - Error messages and warnings - Timing and performance data
stage_files/ - Exported Data
Contains any data marked for export during processing: - Intermediate processing stages (if export=True was used) - Custom export data specified in task configuration - Stage-specific outputs for analysis
derivatives/ - Reports and Visualizations
Quality control reports and visualizations: - Processing reports (PDF/HTML) - ICA component analysis - Data quality visualizations - Before/after comparison plots
Quality Control Assessment#
Processing Summary
Key metrics to evaluate: - Original file information and recording parameters - Channel retention percentage (target: >90%) - Epoch retention percentage (target: >70%) - ICA components removed (typical: 3-8 components)
Data Quality Visualization
Before/after processing comparisons show: - Original EEG traces with artifacts - Cleaned EEG traces post-processing - Frequency domain improvements - Artifact removal effectiveness
ICA Component Analysis
Components are categorized as: - Neural activity (retained) - Eye movement artifacts (removed) - Muscle artifacts (removed) - Cardiac artifacts (removed)
Quality Thresholds#
Acceptable Quality: - Greater than 90% channels retained - Greater than 70% epochs retained - 3-8 ICA components removed - Visible artifact reduction
Review Required: - 85-90% channels retained - 60-70% epochs retained - Less than 3 or more than 12 ICA components removed - Excessive data loss
Flagged Data: - Less than 85% channels retained - Less than 60% epochs retained - Poor artifact removal - Data moved to flagged/ directory
Processed Data Files#
Continuous Data
File: continuous_clean.fif - Artifact-removed continuous EEG - Preserves temporal structure - Suitable for connectivity and time-frequency analysis
Epoched Data
File: epochs_clean.fif - Segmented data with bad epochs removed - Ready for spectral analysis - Optimized for statistical comparisons ICA Solution
The ICA decomposition file contains: - Component weights and topographies - Classification of neural vs. artifact components - Basis for artifact removal decisions
Data Quality Metrics#
Channel Metrics
Channels interpolated: Typically 0-5% of total channels
Channel noise levels: Post-cleaning noise measurements
Bad channel detection: Automated identification of problematic electrodes
Temporal Metrics
Epoch rejection rate: Percentage of data segments removed
Artifact detection: Types and quantities of artifacts identified
Data retention: Amount of usable data remaining after cleaning
Spectral Metrics
Frequency band power across standard EEG bands
Line noise reduction at 50/60 Hz
Spectral quality improvements post-processing
Analysis Considerations#
Quality Indicators
Successful processing typically shows: - Clear artifact component identification in ICA - Reasonable data retention (>70% epochs, >90% channels) - Visible improvement in data quality - Appropriate artifact removal without over-cleaning
Potential Issues
Review data if you observe: - Excessive component removal (>15 ICA components) - Poor data retention (<60% epochs or <85% channels) - Residual artifacts in cleaned data - Over-smoothed or unrealistic signal characteristics
Using Processed Data#
Loading in Python (MNE)
import mne
# Load continuous cleaned data
raw = mne.io.read_raw_fif('bids/continuous_clean.fif')
# Load epoched data
epochs = mne.read_epochs('bids/epochs_clean.fif')
# Perform your analysis
psd = epochs.compute_psd()
Loading Data in MATLAB (EEGLAB):
% AutoClean can export .set files for EEGLAB
EEG = pop_loadset('epochs_clean.set', 'bids/derivatives/');
% Continue with EEGLAB analysis
[spectra, freqs] = spectopo(EEG.data, 0, EEG.srate);
Loading Data in R:
library(eegUtils)
# Load processed data
eeg_data <- import_set("bids/derivatives/epochs_clean.set")
# Continue analysis
psd <- compute_psd(eeg_data)
Quality Control Checklist#
Before proceeding with analysis, verify:
Data Integrity - Processing completed without errors - Output files created successfully - File sizes are appropriate
Quality Metrics - Greater than 70% of epochs retained - Greater than 85% of channels retained - Reasonable number of ICA components removed
Visual Inspection - Clean data exhibits brain-like characteristics - Artifacts successfully removed - Amplitude ranges are physiologically reasonable
Log Review - No critical errors in processing logs - All steps completed successfully - Parameters applied correctly
Troubleshooting Issues#
Excessive Data Loss - Review original data quality - Verify appropriate task selection - Consider parameter adjustments
Poor Artifact Removal - Examine ICA component classifications - Check electrode positioning accuracy - Review preprocessing parameters
Processing Errors - Examine log files in logs/ directory - Verify input data format compatibility - Ensure adequate disk space
Documentation for Publication#
Record the following information: - AutoClean version number - Task name and configuration - Quality metrics and data retention - Custom processing modifications
Example Methods Description: “EEG data were preprocessed using AutoClean v2.0.0 with the RestingEyesOpen task. Data were filtered (1-100 Hz), bad channels interpolated (mean: 2.3%), and artifacts removed using ICA. On average, 78% of epochs were retained after artifact rejection.”
Recommended tutorials: - batch_processing_datasets - Process multiple files efficiently - quality_control_best_practices - Systematic QC procedures - python_integration - Advanced analysis workflows