Saved in:
Bibliographic Details
Main Authors: Liang, Xiao, An, Yuxuan, Wang, Di, Hu, Jiawei, Jiao, Zhicheng, Jing, Bin, Wang, Quan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2512.17213
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917156855218176
author Liang, Xiao
An, Yuxuan
Wang, Di
Hu, Jiawei
Jiao, Zhicheng
Jing, Bin
Wang, Quan
author_facet Liang, Xiao
An, Yuxuan
Wang, Di
Hu, Jiawei
Jiao, Zhicheng
Jing, Bin
Wang, Quan
contents Medical Vision-Language Models (VLMs) are prone to hallucinations, compromising clinical reliability. While reinforcement learning methods like Group Relative Policy Optimization (GRPO) offer a low-cost alignment solution, their reliance on sparse, outcome-based rewards inadvertently encourages models to "overthink" -- generating verbose, convoluted, and unverifiable Chain-of-Thought reasoning to justify answers. This focus on outcomes obscures factual errors and poses significant safety risks. To address this, we propose CheXPO-v2, a novel alignment framework that shifts from outcome to process supervision. Our core innovation is a Knowledge Graph Consistency Reward mechanism driven by Entity-Relation Matching. By explicitly parsing reasoning steps into structured "Disease, Relation, Anatomy" triplets, we provide fine-grained supervision that penalizes incoherent logic and hallucinations at the atomic level. Integrating this with a hard-example mining strategy, our approach significantly outperforms GRPO and state-of-the-art models on benchmarks like MIMIC-CXR-VQA. Crucially, CheXPO-v2 achieves new state-of-the-art accuracy using only 5k samples, demonstrating exceptional data efficiency while producing clinically sound and verifiable reasoning. The project source code is publicly available at: https://github.com/ecoxial2007/CheX-Phi4MM.
format Preprint
id arxiv_https___arxiv_org_abs_2512_17213
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CheXPO-v2: Preference Optimization for Chest X-ray VLMs with Knowledge Graph Consistency
Liang, Xiao
An, Yuxuan
Wang, Di
Hu, Jiawei
Jiao, Zhicheng
Jing, Bin
Wang, Quan
Computer Vision and Pattern Recognition
Machine Learning
Medical Vision-Language Models (VLMs) are prone to hallucinations, compromising clinical reliability. While reinforcement learning methods like Group Relative Policy Optimization (GRPO) offer a low-cost alignment solution, their reliance on sparse, outcome-based rewards inadvertently encourages models to "overthink" -- generating verbose, convoluted, and unverifiable Chain-of-Thought reasoning to justify answers. This focus on outcomes obscures factual errors and poses significant safety risks. To address this, we propose CheXPO-v2, a novel alignment framework that shifts from outcome to process supervision. Our core innovation is a Knowledge Graph Consistency Reward mechanism driven by Entity-Relation Matching. By explicitly parsing reasoning steps into structured "Disease, Relation, Anatomy" triplets, we provide fine-grained supervision that penalizes incoherent logic and hallucinations at the atomic level. Integrating this with a hard-example mining strategy, our approach significantly outperforms GRPO and state-of-the-art models on benchmarks like MIMIC-CXR-VQA. Crucially, CheXPO-v2 achieves new state-of-the-art accuracy using only 5k samples, demonstrating exceptional data efficiency while producing clinically sound and verifiable reasoning. The project source code is publicly available at: https://github.com/ecoxial2007/CheX-Phi4MM.
title CheXPO-v2: Preference Optimization for Chest X-ray VLMs with Knowledge Graph Consistency
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2512.17213