Abstract
Tool-using medical imaging agents must decide when additional evidence is worth requesting and whether a returned image should be used. We study these two decisions in pleural-effusion assessment from chest radiographs. Given a frontal image, a learned request policy conditionally retrieves an available lateral view, after which an image-pair compatibility verifier accepts or rejects the return before two-view re-reading. On MIMIC-CXR-JPG, a counterfactual audit found no measurable mean advantage of a study-matched lateral over a duplicated-frontal control, showing that apparent two-view gains can reflect re-reading and input-format effects. Nevertheless, the learned request policy improved balanced accuracy by 0.77 percentage points while requesting 7.99% of test studies. Under 25% simulated different-subject mis-returns, the compatibility verifier blocked 68.2% of mismatched images while rejecting 32.7% of valid corrective returns. Its terminal effect varied across request policies, motivating future imaging agents that jointly control tool requests and the use of returned evidence.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MedAgent_033.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: https://papers.miccai.org/miccai-2026-sat/supp/MedAgent_033_supp.pdf
Link to Open Review
Open Review Page: https://openreview.net/forum?id=auClkiQKVf
BibTex
@InProceedings{KoYeh_When_MICCAISAT2026,
author = { Ko, Yehyun AND Jung, Junah AND Park, Chang Min AND Lee, Dongheon},
title = { { When to Request and Whether to Trust: A Two-Decision Imaging Agent for Pleural Effusion Assessment } },
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}
}
