Real-Time Wound Edge Detection with YOLOv8 for Robotic Suturing in Resource-Constrained Settings

Abstract

Accurate wound edge detection is a critical step in robotic suturing, especially in settings with limited resources where automation can enhance surgical precision and efficiency. This study presents a real-time computer vision pipeline leveraging the YOLOv8n-seg model to detect and segment wound edges in diverse surgical conditions. The pipeline integrates post-processing techniques, including contour extraction and Ramer-Douglas-Peucker smoothing, to generate precise trajectories suitable for robotic path planning. Trained on a dataset of 2,271 wound images, the model achieved high performance, with mean Average Precision scores ([email protected]) of 0.966 for object detection and 0.961 for segmentation masks, along with high precision and recall. Deployment on an NVIDIA Jetson Nano demonstrated near real-time inference speeds (∼260 ms per image), making the system suitable for interactive robotic applications. This work underscores the feasibility of integrating lightweight, accurate computer vision models into robotic surgical workflows in resource-constrained environments, paving the way for broader adoption of robotic suturing technology. The system specifically responds to infrastructural and workforce constraints common in sub-Saharan Africa. Its standalone, self-contained design eliminates the need for cloud-based processing or expensive server hardware, making its real-time efficiency suitable for rural and under-resourced hospitals. The system design aligns with Human-Computer Interaction (HCI) principles by focusing on usability, autonomy, and contextual relevance for African healthcare environments.