"Weakly Supervised Semantic Segmentation in Histopathology Based on Global Proportions"

"This dataset comprises 325 histopathological images of clear cell renal cell carcinoma (ccRCC) tissue sections, designed to characterize vascular morphology based on CD31 immunohistochemical staining. Each image was scanned at 10\u00d7 magnification and annotated with global proportions for three distinct vascular patterns: high-branching (HB), low-branching (LB), and sinusoidal (SN). The provided annotations include the relative distribution of each vascular class per image. The primary purpose of this dataset is to support the development of weakly supervised semantic segmentation methods, where only global category proportions are available rather than pixel-level annotations. Researchers can use this dataset to build models capable of predicting pixel-level vascular pattern segmentation aligned with the provided global labels. It facilitates advances in computational pathology, particularly in scenarios requiring learning from limited supervision."