Xu Meilong, Hu Xiaoling, Gupta Saumya, Abousamra Shahira, Chen Chao
Stony Brook University, Stony Brook, NY, USA.
Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital and Harvard Medical School, Boston, MA, USA.
Comput Vis ECCV. 2024;15136:271-289. doi: 10.1007/978-3-031-73229-4_16. Epub 2024 Oct 25.
In digital pathology, segmenting densely distributed objects like glands and nuclei is crucial for downstream analysis. Since detailed pixel-wise annotations are very time-consuming, we need semi-supervised segmentation methods that can learn from unlabeled images. Existing semi-supervised methods are often prone to topological errors, ., missing or incorrectly merged/separated glands or nuclei. To address this issue, we propose , the first semi-supervised method that learns the topological representation from unlabeled histopathology images. The major challenge is for unlabeled images; we only have predictions carrying noisy topology. To this end, we introduce a noise-aware topological consistency loss to align the representations of a teacher and a student model. By decomposing the topology of the prediction into signal topology and noisy topology, we ensure that the models learn the true topological signals and become robust to noise. Extensive experiments on public histopathology image datasets show the superiority of our method, especially on topology-aware evaluation metrics. Code is available at https://github.com/Melon-Xu/TopoSemiSeg.
在数字病理学中，分割腺体和细胞核等密集分布的对象对于下游分析至关重要。由于详细的逐像素注释非常耗时，我们需要能够从未标记图像中学习的半监督分割方法。现有的半监督方法往往容易出现拓扑错误，例如腺体或细胞核的缺失、错误合并或分离。为了解决这个问题，我们提出了第一种从未标记的组织病理学图像中学习拓扑表示的半监督方法。主要挑战在于未标记图像；我们只有携带噪声拓扑的预测结果。为此，我们引入了一种噪声感知拓扑一致性损失，以对齐教师模型和学生模型的表示。通过将预测的拓扑分解为信号拓扑和噪声拓扑，我们确保模型学习到真实的拓扑信号并对噪声具有鲁棒性。在公共组织病理学图像数据集上进行的大量实验表明了我们方法的优越性，特别是在拓扑感知评估指标上。代码可在https://github.com/Melon-Xu/TopoSemiSeg获取。
