Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AutocleanEEG ICVision: Automated ICA Artifact Classification Using Vision-Language AI

Zag ElSayed, Grace Westerkamp, Gavin Gammoh, Yanchen Liu, Peyton Siekierski, Craig Erickson, Ernest Pedapati

TL;DR

This work presents ICVision, a vision-language AI that classifies ICA EEG components by analyzing full diagnostic dashboards and generating human-readable explanations, addressing scalability, consistency, and continuity in artifact labeling. By leveraging GPT-4 Vision to interpret multi-panel dashboards, it provides confidence scores and rationale, outperforming traditional feature-based classifiers in expert agreement while preserving neural signals in ambiguous cases. Implemented within the open-source Autoclean platform, ICVision enables scalable, explainable EEG preprocessing and auditability across datasets and sites. The approach demonstrates a paradigm shift from numeric feature classification to image-based, reasoning-enabled artifact decision-making, with tangible benefits for neurodiagnostics and BCI workflows.

Abstract

We introduce EEG Autoclean Vision Language AI (ICVision) a first-of-its-kind system that emulates expert-level EEG ICA component classification through AI-agent vision and natural language reasoning. Unlike conventional classifiers such as ICLabel, which rely on handcrafted features, ICVision directly interprets ICA dashboard visualizations topography, time series, power spectra, and ERP plots, using a multimodal large language model (GPT-4 Vision). This allows the AI to see and explain EEG components the way trained neurologists do, making it the first scientific implementation of AI-agent visual cognition in neurophysiology. ICVision classifies each component into one of six canonical categories (brain, eye, heart, muscle, channel noise, and other noise), returning both a confidence score and a human-like explanation. Evaluated on 3,168 ICA components from 124 EEG datasets, ICVision achieved k = 0.677 agreement with expert consensus, surpassing MNE ICLabel, while also preserving clinically relevant brain signals in ambiguous cases. Over 97% of its outputs were rated as interpretable and actionable by expert reviewers. As a core module of the open-source EEG Autoclean platform, ICVision signals a paradigm shift in scientific AI, where models do not just classify, but see, reason, and communicate. It opens the door to globally scalable, explainable, and reproducible EEG workflows, marking the emergence of AI agents capable of expert-level visual decision-making in brain science and beyond.

AutocleanEEG ICVision: Automated ICA Artifact Classification Using Vision-Language AI

TL;DR

This work presents ICVision, a vision-language AI that classifies ICA EEG components by analyzing full diagnostic dashboards and generating human-readable explanations, addressing scalability, consistency, and continuity in artifact labeling. By leveraging GPT-4 Vision to interpret multi-panel dashboards, it provides confidence scores and rationale, outperforming traditional feature-based classifiers in expert agreement while preserving neural signals in ambiguous cases. Implemented within the open-source Autoclean platform, ICVision enables scalable, explainable EEG preprocessing and auditability across datasets and sites. The approach demonstrates a paradigm shift from numeric feature classification to image-based, reasoning-enabled artifact decision-making, with tangible benefits for neurodiagnostics and BCI workflows.

Abstract

We introduce EEG Autoclean Vision Language AI (ICVision) a first-of-its-kind system that emulates expert-level EEG ICA component classification through AI-agent vision and natural language reasoning. Unlike conventional classifiers such as ICLabel, which rely on handcrafted features, ICVision directly interprets ICA dashboard visualizations topography, time series, power spectra, and ERP plots, using a multimodal large language model (GPT-4 Vision). This allows the AI to see and explain EEG components the way trained neurologists do, making it the first scientific implementation of AI-agent visual cognition in neurophysiology. ICVision classifies each component into one of six canonical categories (brain, eye, heart, muscle, channel noise, and other noise), returning both a confidence score and a human-like explanation. Evaluated on 3,168 ICA components from 124 EEG datasets, ICVision achieved k = 0.677 agreement with expert consensus, surpassing MNE ICLabel, while also preserving clinically relevant brain signals in ambiguous cases. Over 97% of its outputs were rated as interpretable and actionable by expert reviewers. As a core module of the open-source EEG Autoclean platform, ICVision signals a paradigm shift in scientific AI, where models do not just classify, but see, reason, and communicate. It opens the door to globally scalable, explainable, and reproducible EEG workflows, marking the emergence of AI agents capable of expert-level visual decision-making in brain science and beyond.

Paper Structure

This paper contains 23 sections, 8 figures, 2 tables.

Table of Contents

  1. Introduction
  2. The Trifecta of Challenges
  3. ICVision Vision-Language AI
  4. Technical Approach
  5. Vision-Language Reasoning
  6. Verification and Impact
  7. From Engineering to IT and Clinical
  8. Proposed Method
  9. Data Acquisition and ICA Decomposition
  10. Dashboard Generation
  11. Vision-Language Model Pipeline
  12. Model Selection
  13. Prompt Engineering and Input Formatting
  14. Classification Workflow and Output Schema
  15. Auto-Exclusion and Post-Processing
  16. ...and 8 more sections

Figures (8)

  • Figure 1: The System Dashboard ICvision component image labeled parts covering: 1) Topography (Spatial), 2) Time Series (Temporal), 3) Power Spectrum (Frequency), and 4) ERP Image (Trial-level)
  • Figure 2: A logical comparison of Feature-Based vs Vision-Based ICA Classification block diagram
  • Figure 3: The Proposed System Block Diagram
  • Figure 4: Illustration of the configuration example given to the AI thinking machine.
  • Figure 5: Example of the JSON output enables transparent, auditable, and human-aligned classification.
  • ...and 3 more figures