This dataset contains digitized histopathology image data of primary Oral Squamous Cell Carcinoma (OSCC) tumors and corresponding lymph node specimens for semantic segmentation research. The data were curated for the development and evaluation of deep learning–based tumor segmentation models. The dataset includes hematoxylin and eosin (H&E)–stained whole-slide images and corresponding expert-annotated segmentation masks delineating tumor regions. In selected cases, paired cytokeratin (CK) immunohistochemistry slides are provided to support identification of metastatic tumor foci within lymph nodes. All data were anonymized prior to release. The dataset is organized for supervised learning workflows and supports patch-level segmentation model training and evaluation. This dataset was created as part of a graduate research project in biomedical image analysis and deep learning for cancer metastasis detection. File structure, data format specifications, and usage guidelines are provided within the archive.