Research On Key Technologies Of Demonstration Learning Base On Vascular Interventional Surgery Device

Minimally invasive interventional surgery is the most effective and mainstream medical means to treat cardiovascular and cerebrovascular diseases.However,in order to obtain the patient’s blood vessel contour information and feed back the guide wire position information during the operation,X-ray radiation is often required for many times during the operation,so doctors need to wear heavy protective lead clothing to prevent radiation hazards,and protective clothing can not achieve all-round protection of the head and hands.Long years of work will bring serious harm to doctors’ health,Therefore,an interventional surgical robot capable of self intubation is developed to assist doctors in surgery.It can greatlyreduce the harm of X-ray to doctors,and has a very important application value.This paper mainly focuses on the in-depth research on how the interventional surgical robot realizes the teaching and learning of independent wire feeding,multi task migration and other related technologies.First,target detection of guide wire tip is studied on the in vitro blood vessel simulation.By comparing the advantages and disadvantages of target detection algorithms,Yolox detection algorithm is selected to locate the guide wire tip.By conducting wire feeding operation on an in vitro simulated blood vessel model,a surgical image dataset is constructed to train the neural network;The pixel coordinates of the guide wire tip are calculated by Yolox algorithm,and the pose image of the tip is obtained;In addition,BP neural network is trained to learn the mapping relationship between motor current and guide wire contact force,so as to obtain force information in surgery.Through the above research,the robot has obtained good intraoperative feedback and improved the success rate and efficiency of surgery.Secondly,the convolutional neural network and short-term memory network are combined to study and build a teaching and learning model,and realize the simulation of expert surgery action by wire feeding robot.Through the analysis of doctor’s operation demonstration,based on the decomposition and quantification of rotation and propulsion of doctor’s actions,the state of guide wire is obtained by camera in real time,and the sequence data set of construction state and action is collected.The convolutional network is used to analyze the data characteristics,and then it is input to the short-and long-term memory network.The follow-up action is predicted through the calculation of the overall teaching model.Finally,on the self-designed wire feeding device in the laboratory,the self surgical wire feeding experiment on the abdominal aorta vascular model was completed,and good results were achieved.Thirdly,by combining the meta learning(MAML)algorithm with the interventional surgery teaching learning method,the meta teaching neural network model is built using the above teaching learning model as the basic learner.The surgical tasks are divided according to the difference of vessel position and path starting point,and the adaptive ability to different tasks is obtained by collecting data to construct a data set and conducting multi task training on the model through teaching.By comparing the result of meta learning training with the random training model without meta learning,it is proved that its performance is superior to the latter.Finally,an in vitro simulation surgery experiment platform was built,and the in vitro physical model experiments were conducted based on the teaching learning model and the meta learning model studied,which verified the effectiveness and safety of the teaching learning autonomous surgery,and the good generalization ability of the meta learning model among different surgical tasks.