Chen Yuxuan, Li Jiawen, Zhu Lianghui, Xu Yang, Guan Tian, Shi Huijuan, He Yonghong, Han Anjia
Shenzhen International Graduate School, Tsinghua University, Shenzhen, 518055, Guangdong, China.
Department of Laboratory Medicine, Shenzhen Children's Hospital, Shenzhen, 518038, Guangdong, China.
Comput Methods Programs Biomed. 2025 Nov;271:108966. doi: 10.1016/j.cmpb.2025.108966. Epub 2025 Jul 23.
Bone metastasis cancer analysis is a significant challenge in pathology and plays a critical role in determining patient quality of life and treatment strategies. The microenvironment and specific tissue structures are essential for pathologists to predict the primary bone cancer origins and primary bone cancer subtyping. By digitizing bone tissue sections into whole slide images (WSIs) and leveraging deep learning to model slide embeddings, this analysis can be enhanced. However, tumor metastasis involves complex multivariate interactions with diverse bone tissue structures, which traditional WSI analysis methods such as multiple instance learning (MIL) fail to capture. Moreover, graph neural networks (GNNs), limited to modeling pairwise relationships, are hard to represent high-order biological associations.
In this paper, we propose a dynamic hypergraph neural network (DyHG) to overcome conventional graph limitations by connecting multiple nodes via hyperedges. A learnable hypergraph structure is obtained through nonlinear transformation, while a Gumbel-Softmax sampling strategy optimizes patch distribution across hyperedges. An MIL aggregator then derives a graph-level embedding for downstream tasks.
Two large-scale datasets for primary bone cancer origins and subtyping classification are constructed from real-world bone metastasis scenarios. Extensive experiments show that DyHG outperforms state-of-the-art (SOTA) baselines by up to 1.28%, demonstrating its capability to model complex biological interactions and enhance analysis accuracy.
We believe that the proposed DyHG can provide auxiliary diagnostic information for bone metastasis analysis and has potential for clinical application.
骨转移癌分析是病理学中的一项重大挑战，在确定患者生活质量和治疗策略方面起着关键作用。微环境和特定组织结构对于病理学家预测原发性骨癌起源和原发性骨癌亚型至关重要。通过将骨组织切片数字化为全切片图像（WSIs）并利用深度学习对切片嵌入进行建模，可以增强这种分析。然而，肿瘤转移涉及与多种骨组织结构的复杂多变量相互作用，传统的WSI分析方法，如多实例学习（MIL），无法捕捉到这些相互作用。此外，图神经网络（GNNs）限于对成对关系进行建模，难以表示高阶生物关联。
在本文中，我们提出了一种动态超图神经网络（DyHG），通过超边连接多个节点来克服传统图的局限性。通过非线性变换获得可学习的超图结构，同时采用Gumbel-Softmax采样策略优化超边上的补丁分布。然后，一个MIL聚合器为下游任务导出图级嵌入。
从实际的骨转移场景中构建了两个用于原发性骨癌起源和亚型分类的大规模数据集。广泛的实验表明，DyHG的性能比最先进的（SOTA）基线高出1.28%，证明了其对复杂生物相互作用进行建模并提高分析准确性的能力。
我们相信，所提出的DyHG可以为骨转移分析提供辅助诊断信息，并具有临床应用潜力。
