Ultrasound-Guided Localization System for Pericardial Effusion Puncture Robot Utilizing Deep Learning Models

The precision of pericardial effusion puncture surgery is crucial for ensuring patient safety and enhancing surgical outcomes, yet traditional approaches rely heavily on the physician's experience. With the advancement of interdisciplinary medical fields, robots for pericardial effusion puncture surgery guided by ultrasound have shown great potential in improving surgical precision and safety. This paper introduces a dataset of ultrasound images of pericardial effusion, a deep classification network, and a semi-supervised residual network (SSRNet), along with a pericardial effusion localization system for ultrasound-guided puncture robots that integrates these two networks. The experimental results indicate that our proposed system achieves an accuracy of 0.992 in classifying segmentable types of pericardial effusion ultrasound images and a Dice coefficient of 0.9868 for semantic segmentation of pericardial effusion and the heart. The system, combining both models, also reached an accuracy of 97.5% in experiments. Therefore, this research not only provides precise localization for robotic pericardial effusion puncture surgery but also offers insights for future research in this field.