发表机构
University of Tartu; STACC; Better Medicine（塔尔图大学; STACC; Better Medicine）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对CT扫描中像素级标注稀缺问题，提出结合多实例学习与基于Compass模型的解剖过滤的弱监督肾肿瘤分类方法，在内部和外部数据集上验证了有效性，F1达0.83。
AI 中文摘要
用于CT扫描分析的深度学习模型常常受到精确像素级标注稀缺的限制，而这类标注需要放射科医生付出大量努力才能完成。仅使用扫描级标签进行训练虽然减少了标注需求，但引入了新的挑战：低监督比率和大输入体积使模型容易过拟合和进行捷径学习。在本研究中，我们探讨了两种互补的方法来应对这些挑战：多实例学习（MIL）和解剖过滤。MIL将CT体数据划分为2D切片实例，从而能够利用基于ImageNet预训练的高效2D架构，而无需计算量大的3D模型。解剖过滤使用我们自监督的身体部位回归模型Compass，将扫描裁剪到与病理相关的子区域，无需分割掩膜。我们在一个内部数据集（TUH）和两个外部数据集（KiTS23和TCGA-KiRC）上评估了两种MIL框架——基于注意力的MIL（ABMIL）和FocusMIL——用于肾肿瘤分类。我们的最佳模型仅使用扫描级标签即在内部测试集上达到了F1=0.83。我们进一步表明，使用Compass模型进行解剖过滤对于基于嵌入的ABMIL的分布外泛化至关重要，而基于实例的FocusMIL则表现出对分布偏移更强的固有鲁棒性。虽然我们在肾肿瘤上进行了评估，但我们认为这是更广泛的弱监督CT分类流程的概念验证，可适用于其他器官和病理类型。
英文摘要
Deep learning models for CT scan analysis are often limited by the scarcity of precise pixel-level annotations, which require significant radiologist effort to produce. Training on scan-level labels alone reduces annotation requirements but introduces challenges: low supervision ratios and large input volumes make models prone to overfitting and shortcut learning. In this work, we investigate two complementary methods to address these challenges: multi-instance learning (MIL) and anatomical filtering. MIL divides CT volumes into 2D slice instances, enabling efficient 2D architectures with ImageNet pretraining rather than computationally demanding 3D models. Anatomical filtering uses Compass, our self-supervised body part regression model, to crop scans to pathology-relevant subregions without requiring segmentation masks. We evaluate two MIL frameworks - Attention-based MIL (ABMIL) and FocusMIL - on kidney tumor classification across one internal dataset (TUH) and two external datasets (KiTS23 and TCGA-KiRC). Our best models achieve F1 = 0.83 on the internal test set using only scan-level labels. We further show that anatomical filtering with the Compass model is critical for the out-of-distribution generalization of embedding-based ABMIL, while instance-based FocusMIL demonstrates greater inherent robustness to distribution shift. While evaluated on kidney tumors, we consider this a proof-of-concept for a broader weakly supervised CT classification pipeline applicable to other organs and pathologies.
Comments16 pages, 3 figures
