Search papers, labs, and topics across Lattice.
This paper introduces Echo-DM, a novel framework for removing artificial markers from clinical ultrasound images using a conditional latent diffusion network combined with a region-aware fusion module. The approach addresses the limitations of existing methods by enabling mask-free inference while preserving anatomical fidelity, thus reducing shortcut bias in automated analyses. Extensive experiments on the Echo-PAIR dataset reveal that Echo-DM outperforms traditional two-stage baselines in both marker removal and image quality, offering a robust solution for clinical applications.
Echo-DM achieves superior ultrasound marker removal while maintaining anatomical integrity, challenging the effectiveness of existing marker removal techniques.
Clinical ultrasound images often contain artificial markers, such as measurement calipers and text, to assist diagnostic interpretation and comparison. However, these markers can introduce shortcut bias in downstream automated analysis, encouraging deep learning models to rely on marker-related cues rather than clinically meaningful anatomy. Existing marker removal methods are either mask-dependent and vulnerable to error propagation, or mask-free deterministic restorers that may over-smooth ultrasound texture and perturb unaffected background regions. To address these challenges, we present Echo-DM, a framework for ultrasound marker removal via conditional latent diffusion and region-aware fusion. Echo-DM follows a common encoder-diffusion-decoder pipeline, where a DiT-based conditional latent diffusion network performs global restoration and a region-aware fusion module enforces preservation-aware image-space refinement under end-to-end mask-free inference. Building on this fixed core design, we further instantiate Echo-DM-V and Echo-DM-R with VAE-based and RAE-based latent modules, respectively, which demonstrates that the Echo-DM architecture is compatible with diverse latent-module instantiations. Extensive experiments on Echo-PAIR, a large-scale paired clinical ultrasound dataset, demonstrate superior marker removal and strong anatomical fidelity compared with representative two-stage baselines, while providing favorable quality--efficiency trade-offs across deployment settings. Data, code and models will be released at https://github.com/MiliLab/Echo-DM.
