核心概念
Automating hemostasis management through blood segmentation is crucial for improving surgical efficiency and safety.
要約
The content introduces HemoSet, a dataset focused on blood segmentation during surgery to automate hemostasis management. It highlights the challenges in detecting blood in surgical fields and the importance of developing specialized algorithms. The dataset aims to enhance the training of models for autonomous blood suction tools, addressing the precision required for robotic applications. Various segmentation models are benchmarked on the dataset, emphasizing the need for specialized approaches due to unique challenges in surgical scenes. The study provides insights into the difficulties specific to blood detection and encourages further research in automated hemostasis.
I. Introduction:
- Surgeons adapt quickly to visual interference from bleeding during surgeries.
- Hemostasis management automation remains underexplored.
II. Methods:
- Data collection involved a thyroidectomy on a porcine model using endoscopic cameras.
- Annotations were hand-labeled with guidelines ensuring consistency.
III. Results:
- Benchmarking of segmentation models showed varying performance across different metrics.
IV. Discussion:
- Existing general segmentation algorithms struggle with capturing pooling geometry in surgical scenes.
- Specialized datasets like HemoSet are essential for training accurate blood segmentation models.
V. Conclusion:
- HemoSet provides labeled data focusing on macrovascular bleeding, offering opportunities for various automation tasks in surgery.
統計
Our dataset features vessel hemorrhage scenarios where turbulent flow leads to abnormal pooling geometries in surgical fields.
HemoSet Videos 11 Frames 102,616 Labeled Frames 857 FPS 30 Resolution 640×480 Avg. Image Coverage 8.06% Avg. Duration 2m37s Avg. STAPLE Prec. 93.3% Avg. STAPLE Spec. 99.6%
引用
"Automation into hemostasis management would offload mental and physical tasks from surgeons."
"Our dataset will be a valuable resource for researchers working on automatic hemostasis algorithms."