Mechanical Correction Of Manipulator Cone-beam CT And The Research On Calibrators Recognition Of Geometric Calibration

Cone-beam CT system is widely used in clinical departments because of its low radiation dose,high quality of reconstruction image and less scanning time.FDK algorithm is commonly used in image reconstruction of cone-beam CT.Due to the influence of machining accuracy and manual installation,the ideal geometric structure needed by FDK algorithm cannot achieve,which will lead to the geometric artifacts in the reconstructed image.On the basis of the existing geometric calibration methods,the research group proposed an on-line geometric calibration method based on the two-dimensional calibration plates.The imaging object and calibrate phantom are scanned at the same time in the method.The conventional method based on threshold segmentation of bearing ball cannot meet requirements,therefore,it is significant to propose a new segmentation method of bearing ball.The conventional cone-beam CT systems usually realize the inherent scanning mode.In order to break through the limitation of traditional cone-beam CT,the research group has installed the manipulator cone-beam CT system.In the process of installing the cone-beam CT,the actual and ideal position relationships between the two manipulators are not consistent because of the errors of manual installation and machining accuracy.And it is important to calibrate the position mapping matrix of the two manipulators.According to above problems,the paper mainly studies on the following two parts.In the first part,three methods of mechanical correction for manipulators are proposed,which are based on the needle tooling,the ball tooling and calibration phantom,respectively.Based on the geometric structure of the flange,the calibration tools of steel needle and steel ball are designed and manufactured.Two calibration tools separately controlled to be contacted by using teaching devices to respectively control the movement of the two manipulators,therefore,two independent coordinates of two manipulators are connected.Then,the mapping matrix of two coordinates is obtained by solving the overdetermined equation with least square.About the method based calibrating phantom,two independent coordinate systems of manipulators are connected through scanning calibration phantom with the manipulator cone-beam CT.Taking the error sum of squares between actual coordinates and calculated coordinates of bearing balls in coordinate system of source manipulator as the objective function,LM algorithm is used to optimize the rotation and translation parameters of mapping matrix.The influence of the number of corresponding points,number of projections and disturbance error on the correction accuracy is analyzed by the simulation experiment.What is more,the horizontal and vertical splicing experiments of five steel needles are achieved by the manipulator cone-beam CT system.Comparing the results of three correction methods,the method based on calibration phantom had higher accuracy.In the second part,in order to realize the on-line geometric calibration method based on calibration plate,the paper proposed a new method for recognizing bearing ball based on U-Net and Mask R-CNN.The preprocessing of projection image by filtering and enhancement is achieved.Then the mask image of bearing ball is gained by U-Net,and extract the overlapping bearing ball area at the same time.The Mask R-CNN recognize each bearing ball after up-sampling of overlapping bearing ball image.Finally,bearing ball coordinates are divided into three categories by ellipse fitting based on arc segment.The proposed method is verified by the simulation experiment and real experiment of circle scanning,respectively.Comparing the results of segmentation accuracy,geometric calibration and image reconstruction by the proposed method and the conventional segmentation method,the proposed method has better results,which is suitable for the on-line geometric calibration based on calibration plate.The results also proved that the on-line geometric calibration method proposed by the research group has practical application potential and value.