| With the rapid development of technologies such as computers,imaging technology,and medical image processing,image guidance plays a vital role in surgical planning,surgical navigation,and image-guided treatment.For real-time and safety considerations,the US(Ultrasound)has become the primary option for intraoperative imaging guidance.The US has low imaging quality,but its two-dimensional image characteristics lead to significant differences in positioning results in three-dimensional space between users.To obtain three-dimensional clear and accurate image information in real-time for intraoperative image guidance,US-MRI(Magnetic resonance imaging)real-time registration fuses the intraoperative two-dimensional US image and the preoperative three-dimensional MRI image,combining the real-time nature of ultrasound and the three-dimensional clarity of MRI Image characteristics.US-MRI real-time registration is a cross-dimensional,multi-modal,deformable,and real-time complex registration task.It is challenging to solve the above four technical problems simultaneously through a one-time registration.This topic adopts a problem decoupling strategy,introduces preoperative 3D US,preoperatively,performs multi-modal deformation registration of 3D US and MRI;intraoperatively,performs2D-3D real-time registration of preoperative 3D US and intraoperative 2D US.The USMRI real-time registration is decoupled into a combination of multi-modal,deformable registration tasks and cross-dimensional,real-time registration tasks.The main research work and results of this subject are:(1)The image guidance system studied in this paper is intended to be used in an USguided dual-arm robotic needle insertion system.The calibration accuracy of the system determines the puncture accuracy of the system.It directly determines the accuracy of the preoperative US 3D reconstruction and the accuracy of the initial position required for intraoperative real-time registration.This paper first studies the system’s calibration.The system has multi-DOF(multi-degree-of-freedom)end-effectors.The existing calibration methods is designed to fix both needle and probe,and it is not easy to efficiently calibration the needle and probe with multi-DOF.This subject proposes a biplane image-based calibration approach for US-guided dual-arm robotic needle insertion system with multi-DOF end-effectors.(2)This paper is based on the multi-resolution registration framework to preform deformation registration of 3D US and MRI.To obtain 3D US and MRI,it is necessary to segment and reconstruct the US and MRI.At present,the high-precision active contour segmentation needs to manually outline and initial contour that is closer to the segmentation result,and the task is relatively large.This paper proposed semiautomatic segmentation based on sequence,which can automatically obtain the initial contour and improve segmentation efficiency.(3)2D-3D US real-time registration is a cross-dimensional real-time registration task.The existing registration methods can be roughly divided into two categories,registration based on similar optimization and registration based on deep learning.In this paper,the method based on similarity optimization is studied,and the initialization positioning correction strategy is proposed to improve the registration speed.However,the registration speed is far from real-time registration.Therefore,this paper continues to study the registration based on deep learning and proposed a patient-specific registration method based on parameter decoupling.This method explores the patientspecific direction of deep network in the medical image processing,and realizes the2D-3D US real-time registration. |
PDF Full Text Request