Abstract
We introduce MechAgent, an agentic system for translational pathology that integrates active mechanism falsification with multi-scale causal inference. Unlike traditional systems that passively generate post-hoc explanations, MechAgent transforms a multimodal foundation model into an autonomous reasoning engine that constructs, tests, and refines mechanistic hypotheses. The key novelty lies in its intervention-based causal framework, where the agent dynamically decomposes latent representations into conserved biological mechanisms; actively acquires evidence to maximally separate competing hypotheses via information-theoretic selection and performs latent interventions to empirically validate if purported mechanisms causally mediate predictions. A multi-agent committee (Pathology, Toxicology, Translational, Evidence Auditor, Adversarial) operates over a typed hypothesis graph, enabling structured deliberation where contradictions trigger targeted evidence acquisition. This architecture bridges the critical gap between statistical pattern recognition and biologically grounded reasoning, offering a path towards translationally robust medical AI.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MedAgent_042.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=wRnP5BGXQa
BibTex
@InProceedings{PatAri_MechAgent_MICCAISAT2026,
author = { Patra, Arijit AND Dandapath, Iman AND A.S., Anudeep AND Sundaresan, Vaanathi},
title = { { MechAgent: Active Mechanism Falsification for Agentic Cross-Species Medical Reasoning } },
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}
}
