CervSpineNet-CSXA-500: Manual Pixel-Level Annotations of Cervical Spinous Processes for X-ray Segmentation

CervSpineNet-CSXA-500 is a manually annotated cervical spine X-ray segmentation dataset comprising 500 lateral cervical radiographs randomly selected from the publicly available Cervical Spine X-ray Atlas (CSXA). The dataset provides pixel-level segmentation annotations of posterior cervical anatomical structures, mainly spinous processes, developed as part of the CervSpineNet research project. The original binary segmentation masks were created through manual contour tracing using a tablet and stylus, followed by contour processing to generate pixel-level segmentation masks. A subset of the annotations underwent clinical expert review and verification. Additional anatomical-level separate masks (C1–C7) were derived through connected-component analysis and superior-to-inferior ordering, followed by visual verification of the assigned labels. The dataset includes: 500 original CSXA cervical radiographs and their corresponding binary segmentation masks. C1–C7 level-labeled segmentation masks, stored as PNG images. COCO JSON annotations for binary and level-specific segmentation, using lossless run-length encoding (RLE). Image-level metadata and the original training/testing partition (400 training and 100 testing images). Documentation describing annotation methodology, label definitions, data organization, and usage. This dataset supports research in cervical radiograph segmentation, anatomical structure localization, and benchmarking of semantic and instance segmentation methods. It extends the original CSXA dataset with manually generated spinous process segmentation annotations and is associated with the published CervSpineNet study. Source dataset: Ran et al. (2024), Scientific Data. https://doi.org/10.57760/sciencedb.15391 Associated publication: Sawant et al. (2026), Frontiers in Bioengineering and Biotechnology. https://doi.org/10.3389/fbioe.2025.1733689 License: Creative Commons Attribution 4.0 International (CC BY 4.0).