SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2412.06141 · ICML · 2025

MMedPO: Aligning Medical Vision-Language Models with Clinical-Aware Multimodal Preference Optimization

Huaxiu Yao, Peng Xia, Yun Li, Kangyu Zhu, Hongtu Zhu, Sheng Wang

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 17 functions out of this paper's own repositories and ran 7 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
aiming-lab/mmedpo canonical 5 of 15
aiming-lab/MMedPO canonical 2 of 2
FunctionStatusWhere it lives
whitespace_tokenize Ran aiming-lab/mmedpo/curation/Sample_Zero-Shot_Grounding_RSNA/models/tokenization_bert.py
code served (permissive licence) · get_code("cf9ffa02a42184af")
add_gaussian_noise Ran aiming-lab/MMedPO/inference/llava-med-1.5_vqa.py
code served (permissive licence) · get_code("d2bfc515570f0eed")
create_loader_RSNA Ran aiming-lab/MMedPO/curation/Sample_Zero-Shot_Grounding_RSNA/dataset/dataset_RSNA.py
code served (permissive licence) · get_code("c15c37b39ee4184f")
get_chunk Ran aiming-lab/mmedpo/inference/llava-med-1.5_report.py
code served (permissive licence) · get_code("42a46570620cd9fa")
load_vocab Ran aiming-lab/mmedpo/curation/Sample_Zero-Shot_Grounding_RSNA/models/tokenization_bert.py
code served (permissive licence) · get_code("e7fbc7a74a3457c7")
split_list Ran aiming-lab/mmedpo/inference/llava-med-1.5_report.py
code served (permissive licence) · get_code("076c252c52cbb161")
split_to_even_chunks Ran aiming-lab/mmedpo/train/dpo/llava_trainer_weighted.py
code served (permissive licence) · get_code("10893c4608c08075")
add_gaussian_noise Not yet run aiming-lab/mmedpo/inference/llava-med-1.5_report.py
code served (permissive licence) · get_code("8f611a3a3069da33")
bleu Not yet run aiming-lab/mmedpo/eval/eval_report.py
code served (permissive licence) · get_code("cd8fbb4ca9a20927")
generate_noised_image Not yet run aiming-lab/mmedpo/curation/Sample_Zero-Shot_Grounding_RSNA/inference_attention-map_score.py
code served (permissive licence) · get_code("de5de3e13616d4f2")
get_mm_adapter_state_maybe_zero_3 Not yet run aiming-lab/mmedpo/train/dpo/llava_trainer_weighted.py
code served (permissive licence) · get_code("bb35e3ac741bb2c9")
get_tokenizer Not yet run aiming-lab/mmedpo/curation/Sample_Zero-Shot_Grounding_RSNA/inference_attention-map_score.py
code served (permissive licence) · get_code("d070ec2620683428")
load_jsonl Not yet run aiming-lab/mmedpo/eval/eval_report.py
code served (permissive licence) · get_code("77678f2758141df9")
maybe_zero_3 Not yet run aiming-lab/mmedpo/train/dpo/llava_trainer_weighted.py
code served (permissive licence) · get_code("735025744c1ab0cf")
normalize_word Not yet run aiming-lab/mmedpo/eval/eval_report.py
code served (permissive licence) · get_code("8f1a44ab20956301")
score_cal Not yet run aiming-lab/mmedpo/curation/Sample_Zero-Shot_Grounding_RSNA/inference_attention-map_score.py
code served (permissive licence) · get_code("e55695b0b11e1e76")
tensor_to_serializable Not yet run aiming-lab/mmedpo/inference/utils.py
code served (permissive licence) · get_code("af42eb0093076ad3")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

The advancement of Large Vision-Language Models (LVLMs) has propelled their application in medicine. However, Medical LVLMs (Med-LVLMs) encounter factuality issues due to modality misalignment, where the models prioritize textual knowledge over visual input, causing hallucinations that conflict with medical images. Previous attempts on preference optimization have inadequately mitigated clinical relevance in preference data, making these samples easily distinguishable and reducing alignment effectiveness. To address this challenge, we propose MMedPO, a novel multimodal medical preference optimization approach that considers the clinical relevance of preference samples to enhance Med-LVLM alignment. MMedPO curates multimodal preference data by introducing two types of dispreference: (1) plausible hallucinations injected through target Med-LVLMs or GPT-4o to produce medically inaccurate responses, and (2) lesion region neglect achieved through local lesion-noising, disrupting visual understanding of critical areas. We then calculate clinical relevance for each sample based on scores from Med-LLMs and visual tools, and integrate these scores into the preference optimization process as weights, enabling effective alignment. Our experiments demonstrate that MMedPO significantly enhances factual accuracy, achieving improvements over existing baseline methods by averaging 14.2% and 51.7% across the Med-VQA and report generation tasks. Our code are available in https://github.com/aiminglab/MMedPO.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2412.06141")
get_code_for_paper("2412.06141")
have("2412.06141")

Connect an agent — have() is free.