Abstract
Breast cancer care depends on patient–oncologist communication that can address medical concerns while also helping patients manage anxiety. Building conversational agents for this setting is difficult because real consultations are private and available dialogue corpora rarely represent emotional concerns across the care pathway. We present a multi-agent framework that converts patient narratives accessed with permission into evidence-constrained longitudinal conversations and compares support agents through a simulated expressed-anxiety proxy. Quality control retained 275 of 284 generated sessions (96.8%), yielding 1,650 support-agent responses with 74.5% scheduled-fact coverage and 99.1% correct prior-session references. The support agent was adapted using conversations from 21 training patients. The base and adapted agents were then evaluated on 14 patient-disjoint patients under matched scenarios, evidence, seeds, and dialogue length. Of 224 matched pairs across two seeds, 218 met the eligibility criteria. On the secondary final-turn measure, the adapted agent had lower proxy anxiety in 109 pairs and the base agent in 77; 32 were tied. Supportive prompting reduced the mean trajectory proxy from 7.774 to 7.296, whereas uniformly rewriting training responses into a fixed supportive structure increased repetition and worsened simulated outcomes.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MedAgent_045.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=D7U6UAmOV8
BibTex
@InProceedings{GhaNed_Toward_MICCAISAT2026,
author = { Ghafouri, Neda AND Gibson, Jane S. AND Renduchintala, Chaithanya AND Tian, Yu AND Dutta, Aritra AND Bedi, Amrit Singh},
title = { { Toward Anxiety-Reducing Conversational AI Agents for Breast Cancer Care } },
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}
}