MIRA: Medical Image Reflection for Agentic Diagnosis

Source

Source name: arXiv
Source kind: paper
Authors / organization: Shengzhi Wang, Jun Yang, Kai Wu, Xiaozhong Ji, Yiwen Ye, Ziyang Chen, Mingliang Xiong, Wen Fang, Mingqing Liu, Mengyuan Xu, Miaoxuan Shan, Caiyan Liu, Bin He, Qingwen Liu
Venue: arXiv: cs.CV
Published: 2026-08-11
Collected: 2026-08-12
URL: http://arxiv.org/abs/2608.10827v1
PDF: https://arxiv.org/pdf/2608.10827v1
DOI: UNKNOWN
Registry group: academic_apis_and_preprints
Evidence role: canonical_academic_index
Language: en
Country / region: UNKNOWN

Plain-Language Brief

Researchers present MIRA, a medical image diagnosis framework that allows an autonomous agent to search for relevant evidence, verify the relevance and consistency of the acquired evidence, and adapt its tool-use strategy. MIRA improves the accuracy of visual grounding and semantic consistency through a two-stage training strategy: first, a tool-augmented Monte Carlo Tree Search data engine is used to explore diverse diagnostic hypotheses and construct supervised fine-tuning trajectories. Second, reinforcement learning is used to improve decision-making through online reflective principle evolution. MIRA outperforms its Qwen3-VL-8B backbone by 7.44 points on nine medical visual reasoning benchmarks.

What the Document Says

MIRA is a medical image diagnosis framework for autonomous evidence search and reflective verification.
MIRA dynamically invokes image-processing operations and web search while evaluating the relevance and consistency of the acquired evidence.
The training strategy for MIRA includes a tool-augmented Monte Carlo Tree Search data engine and reinforcement learning.
The tool-augmented Monte Carlo Tree Search data engine explores diverse diagnostic hypotheses and constructs supervised fine-tuning trajectories.
Reinforcement learning is used to improve decision-making through online reflective principle evolution.
Failure cases are distilled into candidate principles, and only principles that improve held-out rollout rewards are retained in the reinforcement learning process.
Across nine medical visual reasoning benchmarks, MIRA achieves an average score of 64.73.
MIRA improves its Qwen3-VL-8B backbone by 7.44 points.

Key Concepts Mentioned

Medical Image Diagnosis
Autonomous Agent
Evidence Search
Reflective Verification
Image-Processing Operations
Web Search
Visual Grounding Accuracy
Semantic Consistency
Monte Carlo Tree Search
Reinforcement Learning
Online Reflective Principle Evolution
Failure Cases
Candidate Principles
Held-Out Rollout Rewards
Medical Visual Reasoning Benchmarks

Claims Made by the Document

MIRA improves the accuracy of visual grounding and semantic consistency.
MIRA outperforms its Qwen3-VL-8B backbone.

Evidence Used

Improvement in average score of 64.73 across nine medical visual reasoning benchmarks.
Improvement of 7.44 points in Qwen3-VL-8B backbone.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

Medical visual agents can use tools to inspect images and retrieve external knowledge, but indiscriminate tool use may introduce noisy or misleading evidence. Reliable diagnosis therefore requires not only acquiring additional observations, but also verifying whether tool actions are necessary and whether the resulting evidence supports the current hypothesis. We introduce MIRA (Medical Image Reflection for Agentic Diagnosis), a medical visual diagnostic framework for autonomous evidence search and reflective verification. MIRA dynamically invokes image-processing operations, including zooming, grounding, pointing, rotation, and measurement, as well as web search, while evaluating the relevance and consistency of the acquired evidence. We develop MIRA through a two-stage training strategy. First, a tool-augmented Monte Carlo Tree Search data engine explores diverse diagnostic hypotheses and jointly verifies visual grounding accuracy and semantic consistency to construct supervised fine-tuning trajectories. Second, reinforcement learning further improves decision-making through online reflective principle evolution: failure cases are distilled into candidate principles, and only principles that improve held-out rollout rewards are retained. Across nine medical visual reasoning benchmarks, MIRA achieves an average score of 64.73, improving its Qwen3-VL-8B backbone by 7.44 points. It also increases useful tool-use judgments from 56.2% to 73.8% and reduces harmful judgments from 8.9% to 1.6%. Qualitative analyses show that MIRA can re-examine evidence, correct premature conclusions, and adapt its tool-use strategy. Project page: https://MIRA-VL.github.io/