A large 4 K dataset of supermicrosurgical anastomosis at the 0.2 mm scale

Supermicrosurgery, involving vessels smaller than 0.8 mm, is an important emerging field in reconstructive surgery. Its development is limited by the lack of suitable training resources and the current performance of microsurgical robotic systems. Existing robots provide limited intelligent assistance, and their anastomosis procedures are significantly slower than manual techniques. Here, we present the Microsurgical Silicone Tube Anastomosis dataset (MSTA), a large-scale 4 K video dataset of 0.2-mm silicone tube anastomosis procedures. The dataset includes 452 procedures performed by 571 operators from 158 hospitals across 28 provinces in China, with a total of 40.77 hours of 4 K video (3,840 × 2,160 pixels). All procedures were performed using standardized microsurgical instruments and imaging settings to ensure consistency and comparability. The participants represented a wide range of experience levels, with completion times ranging from 125 to 600 seconds, providing a broad spectrum of technical performance for analysis. Each procedure is accompanied by quantitative metadata, including completion time and expert-assessed performance scores based on predefined criteria. These data enable objective assessment of technical proficiency and analysis of performance-related factors. In addition, the dataset contains 47,452 segmentation annotation surgical frames, supporting computer vision studies. The MSTA dataset addresses the current gap in supermicrosurgical training data, and supports both surgical-robotics development and human training via objective performance assessment. Its scale and standardized design provide a reliable basis for building intelligent assistance systems and establishing benchmarks for skill evaluation in supermicrosurgery.