Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EEG-GPT: Exploring Capabilities of Large Language Models for EEG Classification and Interpretation

Jonathan W. Kim, Ahmed Alaa, Danilo Bernardo

TL;DR

This work investigates EEG-GPT, an LLM-based framework for EEG classification and interpretation that operates effectively with minimal labeled data. By converting EEG features into verbal prompts and fine-tuning a base LLM, it achieves competitive normal/abnormal classification in a few-shot setting and maintains non-trivial zero-shot capability via in-context learning. The approach further demonstrates interpretable, stepwise reasoning through Tree of Thought and access to specialized EEG tools (seizure/spike detectors and qEEG comparison), providing transparent decision-making and potential clinical trust. Although DL models trained on raw EEG still outperform EEG-GPT in some cases, EEG-GPT offers data efficiency and explainability, highlighting a promising direction for trustworthy AI-assisted EEG analysis in clinical contexts.

Abstract

In conventional machine learning (ML) approaches applied to electroencephalography (EEG), this is often a limited focus, isolating specific brain activities occurring across disparate temporal scales (from transient spikes in milliseconds to seizures lasting minutes) and spatial scales (from localized high-frequency oscillations to global sleep activity). This siloed approach limits the development EEG ML models that exhibit multi-scale electrophysiological understanding and classification capabilities. Moreover, typical ML EEG approaches utilize black-box approaches, limiting their interpretability and trustworthiness in clinical contexts. Thus, we propose EEG-GPT, a unifying approach to EEG classification that leverages advances in large language models (LLM). EEG-GPT achieves excellent performance comparable to current state-of-the-art deep learning methods in classifying normal from abnormal EEG in a few-shot learning paradigm utilizing only 2% of training data. Furthermore, it offers the distinct advantages of providing intermediate reasoning steps and coordinating specialist EEG tools across multiple scales in its operation, offering transparent and interpretable step-by-step verification, thereby promoting trustworthiness in clinical contexts.

EEG-GPT: Exploring Capabilities of Large Language Models for EEG Classification and Interpretation

TL;DR

This work investigates EEG-GPT, an LLM-based framework for EEG classification and interpretation that operates effectively with minimal labeled data. By converting EEG features into verbal prompts and fine-tuning a base LLM, it achieves competitive normal/abnormal classification in a few-shot setting and maintains non-trivial zero-shot capability via in-context learning. The approach further demonstrates interpretable, stepwise reasoning through Tree of Thought and access to specialized EEG tools (seizure/spike detectors and qEEG comparison), providing transparent decision-making and potential clinical trust. Although DL models trained on raw EEG still outperform EEG-GPT in some cases, EEG-GPT offers data efficiency and explainability, highlighting a promising direction for trustworthy AI-assisted EEG analysis in clinical contexts.

Abstract

In conventional machine learning (ML) approaches applied to electroencephalography (EEG), this is often a limited focus, isolating specific brain activities occurring across disparate temporal scales (from transient spikes in milliseconds to seizures lasting minutes) and spatial scales (from localized high-frequency oscillations to global sleep activity). This siloed approach limits the development EEG ML models that exhibit multi-scale electrophysiological understanding and classification capabilities. Moreover, typical ML EEG approaches utilize black-box approaches, limiting their interpretability and trustworthiness in clinical contexts. Thus, we propose EEG-GPT, a unifying approach to EEG classification that leverages advances in large language models (LLM). EEG-GPT achieves excellent performance comparable to current state-of-the-art deep learning methods in classifying normal from abnormal EEG in a few-shot learning paradigm utilizing only 2% of training data. Furthermore, it offers the distinct advantages of providing intermediate reasoning steps and coordinating specialist EEG tools across multiple scales in its operation, offering transparent and interpretable step-by-step verification, thereby promoting trustworthiness in clinical contexts.
Paper Structure (15 sections, 7 figures, 2 tables)

This paper contains 15 sections, 7 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Few- and zero-shot Learning
  4. Feature selection and fine-tuning
  5. Few- and zero-shot learning
  6. Evaluation of reasoning capability in EEG tool usage
  7. Tree of Thought
  8. Specialized software tools
  9. Automated seizure detection model
  10. Automated spike detection model
  11. qEEG feature comparison tool
  12. Results
  13. EEG-GPT demonstrates few- and zero-shot learning proficiency
  14. EEG-GPT navigates EEG tool usage with tree-of-thought reasoning
  15. Discussion & Future Work

Figures (7)

  • Figure 1: Diagram depicting the process of fine-tuning large language models bratanic2023diagram
  • Figure 2: Pipeline for few-shot experiment
  • Figure 3: Example prompt-completion pair (formatted for readability)
  • Figure 4: Simplified diagram of clinical epileptologist workflow
  • Figure 5: Resulting AUROCs of normal/abnormal classification task, plotted against proportion of training data used to fit model
  • ...and 2 more figures