Abstract
Multi-site MRI harmonization is essential for mitigating scanner effects, but existing methods rely on different data assumptions and often require researchers to manually select specific pipelines and configure specialized settings. We present a novel HarmoAgent, an autonomous large language model agent for adaptive and traceable multi-site MRI harmonization. HarmoAgent converts natural-language queries into structured task objects, validates data and metadata requirements, and routes each task to an appropriate paired, unpaired, or few-shot harmonization skill. It then executes the selected workflow and returns harmonized outputs with standardized evaluation and provenance records. We evaluated candidate reasoning backbones using expert-validated queries. The selected GLM-5.2 backbone achieved 99.0% routing accuracy and a macro-F1 score of 98.53%. Experiments on traveling-subject dataset further showed that HarmoAgent can orchestrate heterogeneous image-level harmonization skills through a unified workflow. These results suggest that agentic orchestration can reduce manual pipeline configuration and support extensible, consistent, and traceable MRI harmonization.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MedAgent_035.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=IF6uiMJeq8
BibTex
@InProceedings{WanChe_HarmoAgent_MICCAISAT2026,
author = { Wang, Cheng AND Chen, Mingjin AND Zhang, Xin},
title = { { HarmoAgent: An Autonomous LLM Agentic Framework for Adaptive and Traceable Multi-Site MRI Harmonization } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17263},
month = {pending},
page = {pending}
}