The Research On Stent Enhancement And Guidewire Segmentation Algorithm In X-ray Image

X-ray image as an important technical means of adjuvant treatment of cardiovascular disease,which contains stent,guidewire,blood vessels,lesions and other information.The information can be fed back to the doctor in real time through the navigation system of the interventional surgery robot,which can improve the fusion degree between the doctor and the interventional surgery robot,so as to improve the safety of the doctor using the robot to operate the surgery.Due to the strong flexibility of the guidewire,it is difficult for doctors to accurately control the movement of the front end of the guidewire and the transmission of the bracket.Therefore,the real-time positioning and enhanced display of the bracket,and the analysis of the shape and position of the guidewire can help doctors understand the operation status more accurately,operate the guidewire more accurately,and improve the success rate of the operation.This thesis mainly studies the enhancement display algorithm and guidewire segmentation algorithm of stent in X-ray image.At present,there are few researches on stent tracking and enhanced display.The existing researches are mainly based on image domain processing.This method has some limitations and will be affected by the image quality.In this thesis,in the process of stent tracking and positioning,the time sequence information is introduced,the stent is located by balloon point,and the visualization effect of stent is improved by enhancement algorithm.The existing guidewire segmentation algorithm is analyzed from the perspective of image segmentation,ignoring the characteristics of the guidewire itself.In this thesis,we design a loss function for the guidewire segmentation by introducing the prior topological structure of the guidewire.At the same time,due to the difficulty of medical image annotation,this thesis designs a semi supervised learning framework to improve the accuracy of guidewire segmentation with the help of unlabeled data.The main work of this thesis is as follows(1)A real-time X-ray stent detection tracking enhancement algorithm based on time sequence is designed.The algorithm divides stent tracking into two stages.In the preprocessing stage,the suspected balloon point pairs in each frame are detected,and the accurate balloon point pairs are found according to the statistical information in the time sequence.In the second stage of the algorithm,the suspected balloon point pairs of each image are detected in real time.According to the balloon point pair information obtained in the first stage,the most matching balloon point pair is found,and the stent structure on the image is enhanced by registration and superposition.Experimental results show that the algorithm can ensure the reliability and real-time,and improve the visual scoring of stent.(2)A loss function for guidewire segmentation is proposed.Combined with the prior information of guidewire structure,the loss function is divided into two parts: the dice loss function based on distance graph weighting and the collinearity constraint term based on second-order smoothing.The dice function based on distance graph weighting can effectively suppress other similar curve structures,and constrain the segmentation result to be closer to the guidewire label in shape.The collinearity constraint based on second-order smoothing can ensure the topological similarity and coherence of guidewires.The experimental results show that the method can remove the influence of noise and similar structure,and obtain a complete and continuous guidewire structure.(3)A semi supervised guidewire segmentation framework is designed,and an optimization method based on training rounds and data usage is proposed.The semi supervised guidewire segmentation framework uses dual task learning method to solve the problem of guidewire data annotation,introduces level set function prediction of perceptual geometry as auxiliary task,and uses dual task consistency strategy to establish the relationship between labeled data and unlabeled data.In order to describe the role of unlabeled data in the optimization process,this thesis introduces the concept of effective data quantity,designs a weight function,introduces the function based on training rounds and normalized effective data quantity.Experimental results show that semi supervised guidewire segmentation can effectively improve the feature extraction ability and classification ability of the model.