SimuScope: Realistic Endoscopic Synthetic Dataset Generation through Surgical Simulation and Diffusion Models

This is a dataset for the paper entitled "SimuScope: Realistic Endoscopic Synthetic Dataset Generation through Surgical Simulation and Diffusion Models" accepted to IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2025, Tucson, AZ, USA.

Abstract

Cholecystectomy, a common general surgical procedure, serves as the basis for a highly detailed simulation pipeline introduced in this work. The simulator begins with surgical instruments inserted into an inflated abdominal cavity, visualizing key anatomical structures such as the liver, gallbladder, cystic duct, and artery. The simulation replicates critical procedural stages, including dissection of the hepatocystic triangle, achieving the critical view of safety, and subsequent clipping and cutting of the cystic duct and artery, followed by gallbladder removal from the liver bed. At its current development stage, the simulator does not support irrigation or specimen removal functionalities.

The dataset produced by this simulator includes 13,064 images, accompanied by a rich variety of ground-truth annotations: depth and normal maps, optical flow, tool masks, semantic segmentation, and procedure-specific labels like tissue bleeding, damage, and coagulation. This dataset is designed to meet the demanding requirements of modern computer-assisted surgical systems, offering a level of complexity and realism that surpasses existing synthetic datasets. These features ensure that the dataset not only enhances training for surgical systems but also aids in bridging the gap between synthetic and real-world data.

Technical info

The SimuScope dataset includes a collection of 13,064 images generated by the simulator, organized into subfolders: blood_map, color, depth_maps, normal_map, segmentation, and tools_mask. Each subfolder represents a specific aspect of the simulation:

Blood Map: Highlights bleeding tissue to provide real-time feedback on surgical interventions.

Color: Displays the realistic visual appearance of the surgical scene.

Depth Maps: Offers spatial information about the distance of objects within the scene, aiding in spatial understanding.

Normal Map: Delivers detailed surface orientation information to enhance simulation realism.

Segmentation: Categorizes anatomical structures and tools for precise identification during the simulation.

Tools Mask: Identifies and highlights specific surgical instruments in use.

This structure facilitates a comprehensive and detailed analysis of the surgical environment for research and development purposes.

Access priviliges

The SimuScope dataset is publicly accessible to the scientific community through Zenodo. Researchers may use this dataset exclusively for scientific research purposes. Access does not require co-authorship or the formation of a research collaboration.

If you find this dataset useful, please consider citing this paper:

@misc{martyniak2024simuscoperealisticendoscopicsynthetic,      title={SimuScope: Realistic Endoscopic Synthetic Dataset Generation through Surgical Simulation and Diffusion Models},       author={Sabina Martyniak and Joanna Kaleta and Diego Dall'Alba and Michał Naskręt and Szymon Płotka and Przemysław Korzeniowski},      year={2024},      eprint={2412.02332},      archivePrefix={arXiv},      primaryClass={cs.CV},      url={https://arxiv.org/abs/2412.02332}, }