Research on Real-time Lesion Detection Algorithm for Robotic Ultrasound-Guided Images for Percutaneous Transluminal Surgery

With the advancement of robotic technology and ultrasound imaging, robot-assisted percutaneous puncture under ultrasound guidance has become a hot research topic in recent years. Currently, due to issues such as the real-time performance and accuracy of autonomous ultrasound image recognition, robot-assisted percutaneous puncture surgeries largely rely on the experience and professional judgment of physicians. To enhance the autonomy of surgical robots and reduce the operational burden on surgeons as well as training costs for surgical interns, this paper proposes a real-time lesion detection algorithm for robot-assisted percutaneous puncture under ultrasound guidance based on YOLOv8. Firstly, the SimAM (Simultaneous Attention Module) attention mechanism is introduced into the network model to enhance the extraction of lesion features in ultrasound images and the differentiation from the background. Secondly, depthwise separable convolutions replace some traditional convolutions in the feature extraction network to reduce the model's parameter count, further enhancing the detection speed, and the CIoU (Complete IoU loss) is used instead of the original loss function to improve detection accuracy. Finally, a robot-assisted percutaneous puncture system under ultrasound guidance is set up, incorporating the entire real-time lesion detection algorithm. Experimental results show that with the guidance of this algorithm, physicians can more easily complete puncture procedures, achieving a puncture accuracy of 94.43% and a time of 47.27s in multiple puncture experiments, with a real-time frame rate improvement of 11.9 FPS compared to the traditional YOLOv8 algorithm, proving the effectiveness of this real-time lesion detection algorithm for ultrasound-guided percutaneous puncture surgical robots.