ALICE

ALICE automates labeling of independent components in electroencephalography (EEG) recordings using Independent Component Analysis (ICA) to identify non-brain artifacts (e.g., eye movements, muscle) and functional oscillations such as alpha and mu rhythms for improved EEG analysis.

Key Features:

• Crowdsourcing Platform: Integrates crowdsourced visual labeling of independent components (ICs) drawn from public and proprietary EEG datasets to build labeled benchmarks.
• Supervised Machine Learning (ML): Supports training and re-training of supervised ML models on labeled IC subsets to optimize classification performance.
• Consensus Labeling Strategy: Implements a consensus mechanism to combine multiple expert labels and reduce variability in IC classification.
• ICA-based IC characterization: Uses ICA decomposition and IC attributes including time-series estimates, amplitude topography, and spectral power distribution for component description.
• Dynamic database for iterative refinement: Maintains an evolving IC database that updates with new labeled data to enable continuous model improvement.

Scientific Applications:

• Artifact Removal: Identification and removal of non-brain signal artifacts (e.g., eye movements, muscle) from EEG recordings to improve signal quality.
• Functional Oscillation Identification: Automatic detection of neural oscillations such as alpha and mu rhythms for analysis of brain activity patterns.
• Movement Artifact Detection in Specific Populations: Enables targeted ML model optimization for movement artifact detection in populations such as healthy or autistic children.

Methodology:

Decomposition of EEG signals using Independent Component Analysis (ICA); crowdsourced visual labeling of ICs and consensus aggregation of expert labels; supervised ML model training and re-training on labeled IC subsets using IC time-series estimates, amplitude topography, and spectral power distributions; continuous database updates for iterative algorithm refinement.