Development of AI-based surgical skill assessment for laparoscopic training is hampered by a scarcity of publicly available, well-annotated datasets, with existing resources being robotic, simulated, in-vivo or insufficiently annotated and poorly aligned with conventional or low-cost training setups. This work introduces LASK (Laparoscopic Skill and Kinematics), a novel multi-modal peg-transfer dataset captured with synchronised 1280x720 video, an NDI Aurora electromagnetic tracking system providing 6D pose, and jaw angle sensors, comprising 3,725 frames with bounding box annotations for tools and tooltips including a complete 2,680-frame test sequence. Participants comprised 38 low-skill novices (early trainees), 41 medium-skill more experienced trainees and 35 high-skill expert consultant surgeons, of whom 84 were adult surgeons and 30 paediatric surgeons, recruited at the 7th (2023) and 8th (2024) Urology Simulation Bootcamps in Leeds and the 2024 British Association of Paediatric Endoscopic Surgeons annual congress. LASK is reported as the first laparoscopic peg-transfer dataset with synchronised video and comprehensive kinematic data, with preliminary multi-tool detection benchmarks using state-of-the-art models showing greater than 95% accuracy. Intended applications include multi-tool detection and tracking, 3D tool pose estimation including jaw opening from monocular video, objective surgical skill assessment and classification, and investigation of the performance impact of handedness and experience, with a public release planned that adds segmentation masks, further annotated validation videos and baseline benchmarks.
CORE (COnnecting REpositories)
CORE (COnnecting REpositories)