标题： 通过级联双尺度重建，最小高分辨率补丁足以表示全幻灯片图像 显示英文标题

标题： Minimal High-Resolution Patches Are Sufficient for Whole Slide Image Representation via Cascaded Dual-Scale Reconstruction

摘要： 全切片图像（WSI）分析由于其千兆像素规模和稀疏分布的诊断区域而仍然具有挑战性。 多实例学习（MIL）通过将WSI建模为补丁的集合来进行切片级预测，从而缓解了这一问题。 然而，大多数MIL方法强调聚合器设计，而忽视了特征提取阶段的特征提取器的影响，该阶段通常在自然图像上进行预训练。 这导致领域差距和次优表示。 自监督学习（SSL）通过预训练任务在弥合领域差距方面显示出前景，但它仍然主要依赖于通用骨干网络，因此需要将WSIs分割成小补丁。 这不可避免地分割了组织学结构，并生成了冗余且相互依赖的补丁，这反过来会降低聚合器性能并大幅增加训练成本。 为了解决这一挑战，我们提出了一种级联双尺度重建（CDSR）框架，证明每张WSI仅需平均9个高分辨率补丁即可实现稳健的切片级表示。 CDSR采用两阶段的选择性采样策略，从模型基础和语义角度识别最具信息量的代表性区域。 然后将这些补丁输入到局部到全局网络中，通过整合细粒度局部细节与全局上下文信息来重建空间一致的高分辨率WSI表示。 与现有的密集采样或SSL流程不同，CDSR针对效率和形态保真度进行了优化。 在Camelyon16、TCGA-NSCLC和TCGA-RCC上的实验表明，CDSR在下游分类任务中实现了准确率提升6.3%和ROC曲线下面积提升5.5%，每个数据集平均仅使用7,070（总数量的4.5%）个高分辨率补丁，优于在超过10,000,000个补丁上训练的最先进方法。 显示英文摘要

摘要： Whole-slide image (WSI) analysis remains challenging due to the gigapixel scale and sparsely distributed diagnostic regions. Multiple Instance Learning (MIL) mitigates this by modeling the WSI as bags of patches for slide-level prediction. However, most MIL approaches emphasize aggregator design while overlooking the impact of the feature extractor of the feature extraction stage, which is often pretrained on natural images. This leads to domain gap and suboptimal representations. Self-supervised learning (SSL) has shown promise in bridging domain gap via pretext tasks, but it still primarily builds upon generic backbones, thus requiring WSIs to be split into small patches. This inevitably splits histological structures and generates both redundant and interdependent patches, which in turn degrades aggregator performance and drastically increases training costs. To address this challenge, we propose a Cascaded Dual-Scale Reconstruction (CDSR) framework, demonstrating that only an average of 9 high-resolution patches per WSI are sufficient for robust slide-level representation. CDSR employs a two-stage selective sampling strategy that identifies the most informative representative regions from both model-based and semantic perspectives. These patches are then fed into a Local-to-Global Network, which reconstructs spatially coherent high-resolution WSI representations by integrating fine-grained local detail with global contextual information. Unlike existing dense-sampling or SSL pipelines, CDSR is optimized for efficiency and morphological fidelity. Experiments on Camelyon16, TCGA-NSCLC, and TCGA-RCC demonstrate that CDSR achieves improvements of 6.3% in accuracy and 5.5% in area under ROC curve on downstream classification tasks with only 7,070 (4.5% of total) high-resolution patches per dataset on average, outperforming state-of-the-art methods trained on over 10,000,000 patches.