Answers are grounded in this paper and cite their sources.
Spine Endoscopic Atlas-Tissue (SEA-T) is a multicenter dataset for pixel-level multi-class semantic segmentation in endoscopic spine surgery. It contains 3,310 expert-annotated frames from 65 patients (1,663 cervical frames from 39 patients; 1,647 lumbar frames from 26 patients) across three centers, plus 11,155 unlabeled frames. Annotations cover nine classes - upper/lower lamina, ligamentum flavum, dural sac, nerve root, adipose tissue, muscle, disc material, and instrument channel - created in 3D Slicer under multi-stage quality control, with Fleiss' kappa of 0.789. The annotated data are split by center, procedure, and anatomy into Center A training, internal validation, internal test, and held-out sub-cohorts, with Center B and Center C external tests; unlabeled frames are kept separate. Baselines spanning convolutional and foundation-model paradigms - nnU-Net v2, CaraNet, MedSAM2, and the SAM-based MA-SAM baseline - report Dice-based evaluation and reveal center-dependent domain shift. SEA-T is released with native-resolution PNG images, NRRD/MRML masks, RGB overlay images for annotation review, and metadata.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: