Spine endoscopic atlas: an open-source dataset for surgical instrument segmentation

The Spine Endoscopic Atlas (SEA) is a comprehensive dataset specifically designed for the segmentation of surgical instruments used in endoscopic spine surgery (ESS). Given the steep learning curve of ESS and the increasing importance of artificial intelligence (AI) systems in improving procedural accuracy and safety, we believe that our dataset will significantly contribute to the development and enhancement of AI models in this field. The SEA dataset comprises 48,498 images and 10,650 instrument segmentations from real-world ESS, with a focus on instruments commonly used in these procedures, such as bipolars, graspers, drills, punches, dissectors, and scissors. We collected data from two medical centers, including Huazhong University of Science and Technology Union Shenzhen hospital and Huidong Jianqiang hospital, to ensure a diverse representation of instruments and scenarios encountered during surgery. The SEA provides a unique combination of classified and unclassified data, supporting both supervised and semi-supervised learning approaches in instrument segmentation.