DINO-Med3D: Bridging Dimension and Domain Gaps in Volumetric Segmentation via Progressive Adaptation

Researchers have proposed DINO-Med3D, a two-stage framework that adapts the DINOv3 vision model for volumetric medical image segmentation, addressing key barriers that previously prevented its use in three-dimensional clinical scans [1]. The DINOv3 encoder has shown strong semantic discrimination in natural imagery, but its direct application to 3D medical tasks has been blocked by dimension and domain gaps [2]. The new framework, detailed in a paper submitted to arXiv on June 17, 2026, repurposes the pre-trained model through a progressive adaptation process [1][2]. In the first stage, a multi-slice embedding module introduces pseudo-3D context to bridge the dimension gap, while a segmentation proxy task simultaneously adapts representations from natural scenes to the medical domain [2]. Lightweight 3D adapters are then added into the frozen backbone to enforce global inter-slice continuity and improve volumetric understanding [2]. A parallel detail recovery stream preserves high-frequency boundary cues that would otherwise be lost during embedding [2]. Experiments across five public datasets showed the approach outperformed state-of-the-art baselines [2]. The paper appears on arXiv, an open-access repository that hosts electronic preprints across physics, mathematics, computer science, and related fields [6]. As of November 2024, the platform receives about 24,000 submissions per month and has surpassed two million total articles [6]. The work is accessible through arXiv’s abstract page, which includes experimental community tools under the arXivLabs framework [4][5]. arXivLabs, launched in 2020, allows third-party collaborators to build features such as citation explorers and code finders directly on article pages, under guidelines that require adherence to openness, community, excellence, and user data privacy [4][5].