[论文] Where to Look Matters: Rethinking Sub-Volume Sampling in 3D Medical Se...

论文概要

研究领域: CV
作者: Junkai Liu, Le Zhang
发布时间: 2026-10-09
arXiv: 2610.13915

中文摘要

三维医学图像自监督学习在预训练阶段通常依赖随机子体积采样。然而，随机裁剪隐含地将所有空间区域视为同等值得观察，尽管解剖学先验和空间冗余使各区域对表征学习的价值天然不同。为此，我们首先通过受控的裁剪质量操控实验证明：更好的子体积带来更好的下游迁移效果，揭示了观察策略是 3D 医学自监督学习中被忽视的瓶颈。受此启发，我们将子体积采样重新定义为「看哪里」的问题。我们提出了一个即插即用框架，用离线的粗到细探针库替代随机裁剪，由解剖学感知、信息量和多样性三个原则引导。我们实例化了两个变体：轻量变体使用几何和图像统计线索，诊断变体使用冻结视觉特征，以检验轻量的体积内在先验是否足够，或更丰富的语义线索能否可靠地提供额外价值。在多个 3D 医学自监督学习骨干网络和下游任务上的大量实验表明，轻量变体相较于随机采样持续提升了迁移性能。机制分析进一步将这些增益归因于更具解剖学合理性和信息量的观察、减少的空间冗余、更高效的学习动态以及更有结构的表征。

原文摘要

Self-supervised learning on 3D medical images commonly relies on random sub-volume sampling during pretraining. However, random cropping implicitly treats all spatial regions as equally worth observing, despite anatomical priors and spatial redundancy that make regions inherently unequal in their value for representation learning. To this end, we first show through controlled crop-quality manipulations that better sub-volumes lead to better downstream transfer, revealing the observation policy as an overlooked bottleneck in 3D medical SSL. Motivated by this finding, we revisit sub-volume sampling as a where-to-look problem. We propose a plug-and-play framework that replaces random cropping with an offline coarse-to-fine probe bank guided by anatomy-awareness, informativeness, and diversity...

自动采集于 2026-10-14

#论文 #arXiv #CV #小凯