Ultrasound-based Surgery Navigation:Image Segmentation,Calibration And Visualization

Ultrasound-based navigation is a progressively essential technique in diagnostic and interventional medical procedures.The objectives of these procedures are to use images for guiding the surgical tool to the targets within the body and operate them.The navigation task is to track coordinate positions of the image probe and instruments to complete these goals.While the comparison of ultrasound imaging to the others medical imaging,ultrasound imaging have more critical advantages of safety,portability,real-time capability and high temporal resolution.There are three primary challenges of this study.Firstly,we need to guarantee the highly accurate in navigation system so that the navigation scene can present an accurate representation of what’s going on inside the body to the surgeon on the computer screen.Secondly,an accuracy segment the tumor in the blur and the inherent speckle noise found in the medical ultrasound image.Finally,visualization methods combine the available data to present an optimal integrated multimodal scene that displays only the appropriate information to the surgeon in real time.This dissertation provides solutions respectively for each of these tasks:Firstly,a calibration method for the tip of surgical instruments is developed.In view of the sensitivity of the traditional calibration method based on the least-squares(LS)to outliers,an adaptive calibration method based on the recursive maximum entropy(RMC)is proposed in this paper.Experiments show that the RMC method is robust to outliers,and the calibration accuracy of the method is good whether or not outliers exist.Secondly,a global and local correntropy K-mean(GLCK)image segmentation algorithm is proposed by combining the local correntropy and the global correntropy energy.The local correntropy force plays a leading role near the target boundary,which is used to attract the level set function curve to reach the target boundary.While the global correntropy force plays the leading role at a distance from the target boundary.The results show that the GLCK model has improved robustness and image segmentation precision in speckle noise and blurred boundaries image,and the calculation time is also significantly reduced.Finally,an integrated multimodal scene for visualization was proposed.We design the virtual 3D navigation scene that can be viewed from any location and consist of artifact representations of the tracked surgical instrument and the real-time 2D ultrasound image plane.