| Cerebral hemorrhage is a life-threatening disease with high rate of death and disability,which number of cases is increasing year by year worldwide.Cerebral hemorrhage drainage,as a minimally invasive surgical treatment,requires a catheter to be implanted into the hematoma area through a small incision on the skull under the guidance of surgical navigation and then the hematoma can be removed.Existing frame-based or optical surgical navigation systems are expensive,complex and time-consuming,leading to low prevalence of clinical application.This thesis studies the image navigation technology of cerebral hemorrhage drainage based on the economical RGB-D(red,green,blue and depth)camera.The RGB-D camera adopted is Microsoft Kinect v2,which is composed of color camera and ToF(time of flight)sensor.The advantage of this method is that the image and depth information of the operating field scene can be collected at the same time for the positioning of the patient’s head and the tracking of the surgical instruments,but the measurement error is relatively large.Therefore,it is necessary to solve the problems of accurate craniofacial reconstruction of patients and accurate tracking of surgical instruments based on RGB-D cameras.On this basis,the automatic spatial registration method of craniofacial reconstruction and preoperative image is explored to integrate the tracking information of surgical instruments and preoperative image information,as well as displaying the relative positions of surgical instruments and hematoma in real time,so as to realize image guidance of cerebral hemorrhage drainage.Since the error of ToF sensor when measuring the depth of human skin tissue seriously affects the accuracy of craniofacial reconstruction,a depth error compensation method based on diffuse photon density theory model is studied.Firstly,through experimental analysis,it is found that the error of depth measurement is mainly caused by the subsurface reflection of skin tissue under near infrared light.Then,by using the diffuse photon density wave theory and the measurement principle of ToF sensor,the subsurface reflection model of skin tissue to near infrared light is established,and the model is applied to compensate the depth error caused by subsurface scattering of skin tissue.Experimental results show that this method can improve the accuracy of craniofacial reconstruction of RGB-D cameras.Surgical instrument tracking is one of the key navigation techniques,which is used to track the position of the drainage tube in cerebral hemorrhage drainage.In order to avoid ambient light interference and precision loss caused by color to depth image space transformation,infrared markers suitable for ToF cameras are designed based on infrared images.Firstly,the circular infrared marker is made of highly reflective material,and its missing depth value is calculated by depth interpolation algorithm.In order to make use of more depth information,virtual marker and its localization algorithm are introduced,and the position and attitude of the tracking object are output by untraced Kalman filter.Experimental results show that this method has high tracking accuracy and flexibility of marking point arrangement.Spatial registration is one of the key navigation technologies.In this thesis,patients’ craniofacial cloud reconstructed by RGB-D camera is registered with patients’ craniofacial cloud extracted by preoperative image.Aiming at the problems of different disturbance factors,partial overlap and inconsistent density of two point clouds from different sensors,the spatial registration method of craniofacial point clouds based on the guide of facial feature points was studied.Firstly,5 facial landmarks with significant features were selected,and the existing algorithm was used to detect facial feature points from color images in the RGB-D camera space.In the medical image space,the deep learning method was used to extract and locate facial landmarks,and the coarse registration of two point clouds was realized by aligning the two feature point sets.In fine registration,the iterative nearest point algorithm and the weight related to the incident angle of the point cloud are introduced to overcome the disturbance factor of the imaging angle of the RGB-D camera,so as to improve the accuracy and robustness of spatial registration.In order to verify the above key technologies,3D Slicer was used as the platform to develop the image navigation visualization module for cerebral hemorrhage drainage,which integrated the surgical instrument tracking module and spatial registration module,and verified the feasibility and accuracy of navigation in the head model of cerebral hemorrhage and human face target experiments respectively.The results show that the image navigation developed based on the economical RGB-D camera can meet the accuracy requirements of cerebral hemorrhage drainage,and significantly shorten the time required for preoperative spatial registration.In this thesis,the key techniques of image navigation for cerebral hemorrhage drainage based on RGB-D camera are studied in view of the problems existing in image navigation for cerebral hemorrhage surgery.By establishing the depth error model of skin tissue measured by RGB-D camera,a hybrid marker and its tracking method suitable for ToF depth camera are designed,and a craniofacial cloud automatic registration method based on facial feature points is proposed to provide a low-cost,fast and convenient image navigation solution for cerebral hemorrhage drainage.It has great theoretical significance and application prospect. |
