Object Detection of Surgical Instruments Based on YOLOv4
Yan Wang, Qiyuan Sun, Guodong Sun, Lin Gu, Zhenzhong Liu
Abstract
Today, minimally invasive surgery is increasingly used in various operations. Compared with traditional surgery, minimally invasive surgery makes patients less painful and recovers faster after surgery. However, the minimally invasive robotic system may damage surgical instruments or patient organs during the operation. The reason for this situation is the narrow visual space and insufficient tactile feedback. In this paper, we applied a real-time convolutional neural network model based on YOLOv4 to detect surgical instruments during surgery. We selected a public dataset for learning CNN. YOLO' s architecture is applied to the model to detect surgical instruments in real time. Related indicators such as recall and precision were calculated to evaluate the performance of the model.