EEG Preprocessing Pipeline with Automated Comparison

Complete MATLAB toolkit for automated EEG preprocessing using EEGLAB - Designed for hospital EEG data with flexible channel montage support (19-128 channels), featuring comprehensive processing, validation, and comparative analysis.

Quick Start Guide

Prerequisites

• MATLAB (R2018b or later recommended)
• EEGLAB toolbox (latest version)
• ICLabel plugin (required for Model 3 ICA functionality)
• Signal Processing Toolbox (for advanced analysis)

Basic Usage

1. Place EEG files in eeg_files/ folder (supports CNT, EDF, GDF formats)
2. Configure settings at the top of the script (see Configuration Options below)
3. Choose processing model:
   • Model 2: Run model2_eeg_prep.m for basic preprocessing (7 steps)
   • Model 3: Run model3_eeg_prep.m for advanced ICA preprocessing (11 steps)
4. Processed files saved to output/ folder:
   • Model 2: [filename]_model2_preprocessed.edf
   • Model 3: [filename]_model3_preprocessed.edf
5. Processing logs saved to logs/ folder

Model Selection Guide

• Use Model 2 for: Quick preprocessing, baseline processing, when ICA is not required
• Use Model 3 for: Research-grade preprocessing with artifact removal, high-quality clean data requirements
• Use ASR Optimization Tool for: Finding the best ASR parameters for your specific dataset
• Use Comparison Tool for: Validating script vs manual processing, evaluating ICA effectiveness

Directory Structure

Model2-EEG-Pipeline/
├── model2_eeg_prep.m          # Basic preprocessing script (7 steps)
├── model3_eeg_prep.m          # Advanced ICA preprocessing script (11 steps)
├── optimize_asr_parameters.m  # ASR parameter optimization tool
├── compare_models.m           # Automated comparison analysis tool
├── prepare_external_data.m    # External data standardization script
├── eeg_files/                 # Input for all EEG files
├── output/                    # Output for all processed files
├── logs/                      # Processing logs and reports
├── compare_models/            # Comparison analysis folder
│   ├── script/                # Script-processed files
│   ├── manual/                # Manually-processed files
│   ├── results/               # Comparison plots and analysis
│   └── logs/                  # Comparison logs
└── README.md                  # This file

Pipeline Components

Model 2: Basic Preprocessing (model2_eeg_prep.m)

7-step basic preprocessing pipeline without ICA:

1. Import with configurable time range 1b. Clean channel labels (remove reference suffixes like -AA, -Ref)
2. Remove unwanted channels (configurable: reference, EOG, EMG, ECG)
3. Downsample to target sample rate
4. Bandpass filter
5. Clean raw data (artifact removal with ASR)
6. Re-reference to average
7. Save as preprocessed EDF

Model 3: Advanced ICA Pipeline (model3_eeg_prep.m)

11-step advanced preprocessing pipeline with automated ICA artifact removal:

1. Import with configurable time range (Steps 1-4 same as Model 2)
2. Clean raw data (artifact removal with ASR)
3. Add channel locations (Dynamic montage detection: 19/21/25/32/64/128 channels)
4. Run ICA decomposition (Extended Infomax)
5. Classify components using ICLabel
6. Remove artifact components (configurable brain threshold, default ≥70%)
7. Re-reference to average
8. Save as preprocessed EDF

ASR Optimization Tool (optimize_asr_parameters.m)

Systematic parameter testing tool to find optimal ASR settings:

• Tests multiple combinations of Burst and Window criteria
• Balances high data retention with signal quality (SNR, Kurtosis)
• Provides per-file breakdown and actionable recommendations
• Usage: Run optimize_asr_parameters from MATLAB command line

Comparison Tool (compare_models.m)

Automated comparison analysis for systematic evaluation:

• Interactive menu system for easy comparison selection
• Multiple comparison types: Script vs Manual, Model 2 vs Model 3
• Statistical significance testing with p-values
• Enhanced visualizations and reporting

External Data Preparation (prepare_external_data.m)

Utility script to standardize external EEG datasets to the required montage.

• Automatically renames channels (e.g., T7 -> T3) based on map
• Strips channel suffixes
• Saves standardized files to output/ with _prepared.edf suffix

Configuration Options

Each script features a centralized USER CONFIGURATION section at the top.

Processing Parameters

IMPORT_TIME_RANGE_SEC = 180;         % Import first N seconds (set to Inf for all data)
TARGET_SAMPLE_RATE_HZ = 125;         % Target sample rate after downsampling
FILTER_LOW_HZ = 0.5;                 % Bandpass filter low cutoff (Hz)
FILTER_HIGH_HZ = 40;                 % Bandpass filter high cutoff (Hz)

ICA Parameters (Model 3 only)

BRAIN_THRESHOLD = 0.7;               % ICLabel: minimum brain probability to keep component

ASR Parameters

ASR_BURST_CRITERION = 23;            % Threshold for burst removal (lower = more aggressive)
ASR_WINDOW_CRITERION = 0.25;         % Proportion of bad channels to trigger window rejection
ASR_WINDOW_TOLERANCES = [-Inf 7];    % Tolerance range for window rejection

Channel Name Cleaning

STRIP_CHANNEL_SUFFIXES = true;       % Remove reference suffixes from channel names
CHANNEL_SUFFIX_PATTERN = '-(AA|Ref)$';  % Regex pattern for suffixes to remove

Folder Configuration

INPUT_FOLDER = fullfile(pwd, 'eeg_files'); % Input data location
OUTPUT_FOLDER = fullfile(pwd, 'output');   % Output data location
LOG_FOLDER = fullfile(pwd, 'logs');        % Log file location

Channel Removal Configuration

Both models support flexible channel removal configuration:

REMOVE_REFERENCE_CHANNELS = true;    % A1, A2, M1, M2, TP9, TP10, etc.
REMOVE_EOG_CHANNELS = true;          % VEOG, HEOG, EOG1, EOG2, etc.
CHANNELS_TO_NEVER_REMOVE = {'Fp1', 'Fp2', 'F3', ...}; % Safety protection

Feature Documentation

ASR Parameter Optimization

The optimize_asr_parameters.m tool runs a sweep of parameter combinations to find the best balance for your specific data.

Configuration:

BURST_VALUES = [20, 30, 40, 50];           % Values to test
WINDOW_VALUES = {0.25, 0.5, 0.75, 'off'};  % Values to test
TARGET_RETENTION = 0.90;                   % Target data retention (e.g., 90%)
MAX_FILES_TO_TEST = 10;                    % Test subset of files (or 'all')

Output:

• Results table with retention, SNR, kurtosis, and quality scores
• Per-file breakdown showing which files pass/fail
• Recommendation for optimal settings
• CSV report saved to logs/asr_optimization_results.csv

Tip

The tool's recommendation is a statistically optimal starting point. We recommend checking values slightly above/below the recommendation (e.g., Burst +/- 1) using visual inspection in EEGLAB, as manual validation may yield even better results for specific noise profiles.

Comparison Analysis

Execute compare_models.m to start the interactive tool.

1. Set up comparison folders:
   • compare_models/manual/model2/ & model3/ - Manually-processed files
   • compare_models/script/model2/ & model3/ - Script-processed files
2. Select comparison type from the menu.
3. View Results: Saved in compare_models/results/ with unique filenames.

Analysis Features:

• Signal quality assessment: Amplitude, variability, SNR improvements
• Power spectral density analysis: Frequency band changes
• Statistical significance testing: Wilcoxon signed-rank tests
• Channel correlation analysis: Signal preservation assessment

Quality Metrics & Thresholds

Processing Quality Validation

• Brain Component Retention: ≥70% brain probability threshold (Model 3)
• Power Change Validation:
  • Normal: <15 dB change
  • Substantial: 15-30 dB change
  • Warning: >30 dB change (likely error)
• Signal Reduction Warnings: Amplitude >20x or Variability >200x reduction flagged

ICA Quality Assessment (Model 3)

• Component classification confidence: ICLabel probability scores
• Retained vs removed component breakdown: Detailed count of artifact types
• Component separation quality: Information entropy measures for classification clarity

Support & Troubleshooting

Common Issues

• Dynamic channel location assignment: Supports 19-128 channel montages with automatic fallbacks
• ICLabel requires channel locations: System automatically selects appropriate montage file
• Memory issues: Reduce IMPORT_TIME_RANGE_SEC if running out of RAM
• GUI not showing: Set SHOW_EEGLAB_GUI = true to debug visually

Error Recovery Features

• Graceful failure handling: Failed files are logged and skipped
• Memory management: Automatic cleanup (onCleanup) prevents memory leaks
• Log file safety: Logs are guaranteed to close even on error

Validation Workflow

1. Process test files with both Model 2 and Model 3
2. Run comparison analysis to validate effectiveness
3. Review quality metrics in generated logs
4. Adjust thresholds based on findings using the Optimization Tool