标题： TRUST：面向跨设备适应的令牌驱动超声风格迁移 显示英文标题

标题： TRUST: Token-dRiven Ultrasound Style Transfer for Cross-Device Adaptation

摘要： 超声图像从不同设备获取时表现出多样的风格，导致下游任务性能下降。 为了缓解风格差异，无配对图像到图像（UI2I）翻译方法旨在将来自源域的图像（对应新设备采集的图像）转移到目标域，其中冻结的任务模型已针对下游应用进行训练。 然而，现有的UI2I方法并未明确考虑过滤最相关的风格特征，这可能导致翻译后的图像与下游任务的需求不匹配。 在本工作中，我们提出TRUST，一种基于标记的双流框架，在保留源内容的同时转移目标域的通用风格，确保内容和风格保持分离。 考虑到目标域中存在多种风格，我们引入了一个基于标记的（TR）模块，从两个角度操作：（1）数据视角——为每个源标记选择“合适”的目标标记，（2）模型视角——在行为镜像损失的指导下，为下游模型识别“最优”的目标标记。 此外，我们在源编码器中注入辅助提示，以使内容表示与下游行为相匹配。 在超声数据集上的实验结果表明，TRUST在视觉质量和下游任务性能方面均优于现有的UI2I方法。 显示英文摘要

摘要： Ultrasound images acquired from different devices exhibit diverse styles, resulting in decreased performance of downstream tasks. To mitigate the style gap, unpaired image-to-image (UI2I) translation methods aim to transfer images from a source domain, corresponding to new device acquisitions, to a target domain where a frozen task model has been trained for downstream applications. However, existing UI2I methods have not explicitly considered filtering the most relevant style features, which may result in translated images misaligned with the needs of downstream tasks. In this work, we propose TRUST, a token-driven dual-stream framework that preserves source content while transferring the common style of the target domain, ensuring that content and style remain unblended. Given multiple styles in the target domain, we introduce a Token-dRiven (TR) module that operates from two perspectives: (1) a data view--selecting "suitable" target tokens corresponding to each source token, and (2) a model view--identifying ``optimal" target tokens for the downstream model, guided by a behavior mirror loss. Additionally, we inject auxiliary prompts into the source encoder to match content representation with downstream behavior. Experimental results on ultrasound datasets demonstrate that TRUST outperforms existing UI2I methods in both visual quality and downstream task performance.