Study On Robot-assisted Medical Ultrasound Scanning System

Ultrasound imaging is one of the most widely used routine examinations in medical practice.However,there are currently multiple issues impacting clinical practice,including a shortage of highly trained sonographers,work related musculoskeletal disorders,high dependency on skills and experience,and the lack of accurate spatial data during free-hand scanning.The application of ultrasound imaging in image-guided therapy is therefore limited.It is imperative to develop novel approaches to increase automation of the ultrasound scanning work flow.In this work,we built a semi-automatic robotic arm assisted ultrasound scanning system.Our system aims to relieve sonographers of repeated scanning motions that causes fatigue and mechanical stress,using a robotic arm.We also aimed to further enhance reproducibility and standardization of the scanning procedure.Explioting the precision and traceablility of the robot arm,we also aimed to make application of ultrasound imaging in the planning and navigation of image-guide minimally invasive therapy.We conducted a thorough study of common ultrasound scanning procedures in a clinic,explored specific tasks that can be automated with high feasibility and addresses practical clinical needs and translated it into specific goals of this project.Our system mainly consists of three functional modules: a stereo visual sensor for guidance,motion planning and control for robotic arm and real-time ultrasound image acquisition.Our proposed system constructs guidance information from a stereo visual sensor,controls robotic arm to hold ultrasound probe for scanning in target area,and captures the 2D ultrasound images along the scanning path.Considering the specific goals based on clinic needs,we proposed the following research plan.Firstly,compared with traditional systems which manage each hardware modules separately,we utilize ROS as the system core to facilitate the complex communication and coordination between different modules,and thus enhance the response time and performance,ease of development and maintainance and scalability of the system.Secondly,in the guidance module,we obtained depth image of the workspace and performed intrinsic and extrinsic calibrations between the depth camera and RGB camera,a RGB-encoded point cloud to represent objects in the workspace was produced.To process the multi-dimension sparse point cloud data efficiently,we customized a Kd-tree based indexing algorithm to efficiently eliminate noisy data and reconstruct surface normal vectors.A semi-automatic scanning ROI localization module was developed to extract a final scanning path for guidance.Thirdly,our system utilizes a light-weight robotic arm UR5 with 6 degrees of freedom for accurate execution of the scanning path.To ensure the time efficiency and smoothness in scanning,path planning and control parameters for each robotic joint was calculated with an efficient inverse kinematics algorithm.Surface normal vectors along the planned path were used to modulate the pose of the ultrasound scanhead for enhanced smoothness during the scan motion.Finally,a 3D reconstruction was perform to demonstrate with a contiguous volumetic scan larger than that achieved by 3D ultrasound scanner or tracking based reconstruction solutions.Different experiments were designed to evaluate the performance of our system.We evaluated the performance of KinectV2 camera by comparing its length measurements on objects against physical measurements.Results showed that,KinectV2 presents smaller error with shorter working distance.Therefore,in our final system,KinectV2 wa set to work at a range of 1.5 m,which can limit the absolute measurement error to 2 mm and relative error to 0.5%.Thus the requirement on guidance can be satisfied.A custom-designed needle was mounted on the robot arm and moved to landmarks located from the KinectV2-based guidance data.With which,the accuracy of localization of the overall system was determined.After an error compensation method was implemented,a localization error smaller than 4 mm was achieved.To provide a proper scanning speed for reference,the impact of different arm velocities in affecting the ultrasound image quality was also determined.The accuracy,efficiency and stability of the overall system was determined by conducting experiments on several phantoms,and human volunteers.The results show that,based on the ROS operation system,our system can extract effective guidance information from KinectV2,then control the UR5 robotic arm with ultrasound probe to move to the target area with satisfactory speed,accuracy and stability,and finally complete the ultrasound image acquisition in an automated manner.The 3D reconstruction results for large targets generated from the consecutive 2D frames also verifies the promising potentials of our robotic ultrasound imaging system.