Focus on What Matters: Enhancing Medical Vision-Language Models with Automatic Attention Alignment Tuning
Published in ACL 2025, 2025
Medical Large Vision-Language Models (Med-LVLMs) often exhibit suboptimal attention distribution on visual inputs, leading to hallucinated or inaccurate outputs. Existing mitigation methods primarily rely on inference-time interventions, which are limited in attention adaptation or require additional supervision. To address this, we propose A3Tune, a novel fine-tuning framework for Automatic Attention Alignment Tuning. A3Tune leverages zero-shot weak labels from SAM, refines them into prompt-aware labels using BioMedCLIP, and then selectively modifies visually-critical attention heads to improve alignment while minimizing interference. Additionally, we introduce a A3MoE module, enabling adaptive parameter selection for attention tuning across diverse prompts and images. Extensive experiments on medical VQA and report generation benchmarks show that A3Tune outperforms state-of-the-art baselines, achieving enhanced attention distributions and performance in Med-LVLMs.
Recommended citation: Chang, Aofei, Le Huang, Alex James Boyd, Parminder Bhatia, Taha Kass-Hout, Cao Xiao, and Fenglong Ma. (2025). "Focus on what matters: Enhancing medical vision-language models with automatic attention alignment tuning." In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics. 1: 9357–9372. https://aclanthology.org/2025.acl-long.460.pdf