用少量目标域图像，让医疗大模型更好适应新场景。
MFM-DA: Instance-Aware Adaptor and Hierarchical Alignment for Efficient Domain Adaptation in Medical Foundation Models
- 设计动态实例感知适配器和层级对齐机制
- 仅用少量无标签目标图像即提升分割精度
- 适合医疗图像领域迁移部署的轻量级方案
医学基础模型（MFMs）在大规模数据上训练后，在多种任务中表现出色，但在实际应用中仍面临域差距问题。即使在源域数据上微调后，模型在目标域的表现依然不佳。为此，我们提出一种少样本无监督域适应框架MFM-DA，仅利用少量无标签目标域图像。方法首先训练一个去噪扩散概率模型（DDPM），再通过提出的动态实例感知适配器与分布方向损失将其适配至目标域，实现源域图像到目标域风格的转换。生成的目标域图像随后输入医学基础模型，引入通道-空间对齐低秩适配（LoRA）以确保特征有效对齐。在视杯与视盘分割任务上的大量实验表明，MFM-DA优于现有先进方法。本工作为真实场景下医学基础模型的域差距问题提供了实用解决方案。代码将公开。
原文摘要 · Abstract (English)
Medical Foundation Models (MFMs), trained on large-scale datasets, have demonstrated superior performance across various tasks. However, these models still struggle with domain gaps in practical applications. Specifically, even after fine-tuning on source-domain data, task-adapted foundation models often perform poorly in the target domain. To address this challenge, we propose a few-shot unsupervised domain adaptation (UDA) framework for MFMs, named MFM-DA, which only leverages a limited number of unlabeled target-domain images. Our approach begins by training a Denoising Diffusion Probabilistic Model (DDPM), which is then adapted to the target domain using a proposed dynamic instance-aware adaptor and a distribution direction loss, enabling the DDPM to translate source-domain images into the target domain style. The adapted images are subsequently processed through the MFM, where we introduce a designed channel-spatial alignment Low-Rank Adaptation (LoRA) to ensure effective feature alignment. Extensive experiments on optic cup and disc segmentation tasks demonstrate that MFM-DA outperforms state-of-the-art methods. Our work provides a practical solution to the domain gap issue in real-world MFM deployment. Code will be available at here.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。