| Medical ultrasound diagnosis is widely used in clinical practice due to its many advantages such as no radiation,low cost,and multiple diagnostic sites.However,the development of ultrasound discipline in China is extremely unbalanced in terms of geography.Some doctors are often in a high-intensity work state,and the use of robots for autonomous ultrasound scanning has become a research hotspot in the field of medical equipment.However,most of the existing autonomous ultrasound robots are developed from the perspective of robot technology,ignoring the learning process of clinical doctors’ ultrasound scanning.This article proposes an autonomous ultrasound scanning method based on ultrasound image and tactile data guidance by introducing tactile sensing data and imitating the scanning action of doctors adjusting the probe through hand touch.The main work and conclusions completed in this article are as follows:Firstly,in order to monitor and describe the contact state between the ultrasonic probe and the kidney model during the scanning process,a tactile sensor suitable for ultrasonic probe scanning was designed.The working principle of tactile sensor was analyzed.The structural design and circuit design were completed.The contact force calibration of tactile sensor installed on the probe was carried out,including data acquisition process of contact force calibration,regression fitting method and experimental effect of contact force prediction.The results show that the root mean square error between real contact force and predicted contact force is 1.639 N,which meets the accuracy requirements of contact force for ultrasonic scanning tasks.Secondly,an ultrasound scanning skill learning network that integrates ultrasound image and tactile data was designed,which specifically includes image autoencoder and pose prediction model.Using the integrated device of ultrasonic probe and tactile sensor designed in this article and Vicon system’s coordinate positioning function,various information collection during teaching process was realized.Data set preparation and model training were completed to verify the fitting effect of ultrasound scanning skill learning model proposed in this article: The average pixel deviation value between autoencoder decoding image and original image is about 16.758,which has high similarity,indicating that the image features extracted by encoding process are effective;The average position error predicted by pose prediction model does not exceed 20% of maximum position adjustment amount,which is consistent with teaching data as a whole.Finally,an autonomous ultrasound robot motion control scheme based on pose increment was proposed.Based on UR5 manipulator arm,an experimental platform for ultrasound robot was built.The kinematics of ultrasound robot were derived,and ROS-based motion control system for ultrasound robot was implemented.Tactile sensor contact force motion compensation experiment and five autonomous ultrasound robot scanning experiments were carried out based on ultrasound robot experimental platform.Among them,motion compensation experiment analyzed the motion state and contact force of robot’s five stages to verify the effectiveness of displacement compensation;Autonomous ultrasound robot scanning experiment recorded the motion state,trajectory and ultrasonic image of robot.Through analysis of robot’s motion process and final results,feasibility of autonomous scanning by ultrasound robot guided by ultrasound image and tactile data proposed in this article was verified. |
