Agentic EEG Pipeline Synthesis Using Large Language Models and the Model Context Protocol.

Authors

Kylie K Higashionna, Jasmine M Meissner, Katy Tarrit

Published in

Sensors (Basel, Switzerland). Volume 26. Issue 20. Oct 13, 2026. Epub Oct 13, 2026.

Abstract

Electroencephalography analysis typically requires multistage processing pipelines for which configuration depends on recording characteristics and the intended analytical objective. Although automated pipelines improve reproducibility, they generally implement predefined processing sequences. We present an agentic framework in which a large language model (LLM) orchestrator synthesizes and executes task-specific EEG processing workflows by selecting, parameterizing, and ordering established signal processing components accessed through Model Context Protocol interfaces. Computational EEG processing and analysis operations are performed using established EEGLAB (version 2026.0.0) implementations running in MATLAB (R2026a), while structured execution results and quality control information are returned to the orchestrator to support adaptation during execution. The framework is evaluated by examining tool-layer execution fidelity, independent pipeline synthesis, and the quality of the resulting EEG output relative to the established DISCOVER-EEG workflow. In an evaluation using eight resting state EEG recordings from four participants, mean epoch retention was 0.845 for the agent-generated pipeline and 0.843 for DISCOVER-EEG, while mean spectral entropy was 4.87 and 4.84 bits, respectively. Mean signal-to-noise ratio was 13.92 dB for the agent-generated pipeline and 11.14 dB for DISCOVER-EEG; however, this difference should be interpreted cautiously because the pipelines used different upper filter boundaries. A secondary analysis examined whether EEG processed using the generated pipeline retained features relevant to Alzheimer's disease versus healthy control classification. These findings demonstrate the feasibility of coupling LLM-based orchestration with EEG processing software to generate executable task-specific workflows with structured execution records and provenance. Further evaluation across larger datasets, independent synthesis runs, and additional EEG paradigms is required in order to establish repeatability and generalizability.

PMID:
42944335
Bibliographic data and abstract were imported from PubMed on 27 Oct 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.