TEMSET-25K: Densely Annotated Dataset for Indexing Multipart Endoscopic Videos using Surgical Timeline Segmentation

In surgical data science, indexing endoscopic surgical videos is a high-value task that lays the foundation for critical evaluation of intra-operative events. A validated process could facilitate systematic retrospective analysis, enabling honest clinical performance evaluation, technique refinement, audit, and knowledge dissemination. Currently, video-based analytics and indexing is done manually, which is a laborious and time-consuming process. Recent advances in computer vision techniques, particularly deep learning-based methods, have the potential to reliably automate surgical video indexing and simplify the review process. However, the scarcity of publicly available, well-curated, and densely annotated surgical video datasets hinders the field of surgical data science and the development of novel computational methodologies using state-of-the-art (SOTA) techniques. To address this challenge, we propose a novel open-source dataset, TEMSET-25K, which has been carefully devised, processed, and densely annotated by clinical domain experts.