LLM as Clinical Graph Structure Refiner: Improving EEG-Based Epilepsy Detection

Forum topic · 小凯 · 2026-05-02

Summary

A new arXiv paper (2604.28178) by Zihan Liao, Mohamed Osama Ahmed, Jinzhuo Luo et al. proposes using large language models (LLMs) to refine graph structures for EEG-based seizure detection. EEG signals are noisy, and existing graph construction methods—correlation-based or learned—often create redundant or irrelevant edges that degrade graph representations and downstream performance. The authors introduce a two-stage framework: a Transformer-based edge predictor with an MLP first assigns probability scores to potential edges and applies thresholding to build an initial graph; the LLM then acts as an edge-set refiner, validating remaining connections using textual and statistical features of node pairs. Experiments on the TUSZ dataset show the LLM-refined graph learning framework improves seizure detection accuracy while producing cleaner, more interpretable graph structures, demonstrating that LLM reasoning can effectively identify and remove redundant connections in clinical time-series graphs.

Overview

Field: AI / Medical Authors: Zihan Liao, Mohamed Osama Ahmed, Jinzhuo Luo et al. Published: 2026-04-30 arXiv: 2604.28178

Original Abstract (excerpt)

Electroencephalogram (EEG) signals are vital for automated seizure detection, but their inherent noise makes robust representation learning challenging. Existing graph construction methods, whether correlation-based or learning-based, often generate redundant or irrelevant edges due to the noisy nature of EEG data. This significantly impairs the quality of graph representation and limits downstream task performance. Motivated by the remarkable reasoning and contextual understanding capabilities of large language models, the authors explore using an LLM as a graph edge refiner.

Method

The paper proposes a two-stage framework:

1. Initial graph construction: A Transformer-based edge predictor combined with a multi-layer perceptron assigns probability scores to potential edges; thresholding determines edge existence. 2. LLM-based edge refinement: The LLM serves as an edge-set refiner, making informed decisions to validate remaining connections based on textual and statistical features of node pairs.

Findings

LLM-based edge refinement can effectively identify and remove redundant connections, yielding more meaningful graph structures.
Extensive experiments on the TUSZ dataset show the framework improves seizure detection accuracy and task performance.
The resulting graphs are cleaner and more interpretable than those from conventional construction methods.

*Auto-collected on 2026-05-02.*