Systeme de visualisation pour l'assistance des chirurgies minimalement invasives du rachis.
详细信息
文摘
Scoliosis is a complex three-dimensional deformity of the spine that may require surgical treatment. Minimally invasive surgeries, as opposed to conventional surgeries, are done by performing small incisions through which surgical instruments and an endoscopic camera can be inserted within the body. This type of surgery offers numerous known benefits, such as reduction of blood loss, of post-surgery pain, of intensive care unit and hospital stay lengths, of complications, of recovery time and of overall treatment costs. However, it brings new challenges for surgeons. Indirect visualization of surgical sites via a monitor that shows images from the endoscopic camera can be disorienting. Depth perception is lost as the view is monocular and the camera lens is very close to the surgical site. This lens proximity also causes a loss in perception of the global spine shape since only a small part of it is visible at one time. These difficulties make minimally invasive surgeries very complex procedures, especially in the case of spine surgeries, where several critical organs are located near the surgical site. Image-guided surgical assistance tools are increasingly available in various fields such as cardiology, neurology and orthopaedics. For spine surgeries, ionizing imaging modalities such as computed tomography or fluoroscopy, involving radiation for both the patient and the surgical team, are often used. These surgical navigation systems are also generally designed for conventional surgeries requiring long incisions to facilitate access to targeted structures. Thus, metal markers can be screwed into specific vertebral levels, a procedure that is both invasive and hard to perform in a minimally invasive context. Recent minimally invasive surgical assistance approaches try to provide depth cues to the surgeon using only the content of endoscopic images. Methods using a stereoscopic endoscope give unsatisfactory results due to the weak disparity between the two acquired views and to the complexity of feature matching in the two views, caused by the presence of smoke, blood and specular reflections. By the same token, approaches based on shape-from-shading algorithms give, to date, acceptable results in controlled environments but do not perform as well with real images. Hence, this research projects objective is to propose a 3D visualization tool for minimally invasive spine surgery assistance. This tool realizes multimodal fusion of the images acquired from the endoscopic camera with a preoperative 3D model acquired by MRI that has been previously segmented and then registered in a coordinate system linked to the surgical site, in order to display more information than what is available in the endoscopic images alone. To achieve this objective, we first develop a model of endoscopic image formation from models available in the literature. We choose a perspective projection camera model for this purpose. However, since the endoscopic camera head can be rotated relative to the endoscope tube in order to explore the body cavity, this additional degree of freedom must also be modeled. To do so, we use a variant of Yamaguchis method to model this extra transformation, while a variant of Wus method allows us to simplify Yamaguchis model without losing any flexibility and to obtain the rotation angle without using a rotary encoder. Then, we develop an offline calibration process for the model parameters. It is based on the calibration algorithm developed by Zhang as implemented in the OpenCV library. We propose a fully automatic identification of calibration object grid points to minimize the need for user intervention. The calibration object is designed to be simple and inexpensive to produce, in addition to allowing high precision and avoiding the need to measure it using a 3D scanner. Next, we propose a method for online update of a subset of parameters that vary during surgery, namely the cameras extrinsic parameters and the relative rotation angle between the camera head and the endoscope tube. This method relies on the use of an optical tracking system, the MicronTracker by Claron Technology Inc. This system is able to track in real time the position and orientation of markers which can be easily designed and produced. One such marker is placed on the endoscope tube to track in real time the cameras location, while another is placed on the camera head to measure its rotation angle with respect to the tube, as proposed by Wu. Subsequently, we propose two visualization modes to efficiently display the combined information from endoscopic images and the preoperative 3D model. The first of these is an augmented reality visualization mode in which reality, i.e. the endoscopic images, is augmented with information from the preoperative model such as its projection on the images. The second mode is an augmented virtuality visualization in which the preoperative model acquired from MRI is augmented by adding reality, i.e. the endoscopic images, which must be properly positioned according to the endoscopes location relative to the spine. We validate different parts of the calibration method using simulations as well as experiments with real endoscopic images. Simulations, in which the calibration data is artificially noised, show that our method is tolerant to noise. Validation experiments on real images allow us to demonstrate the proposed image formation model as well as results reproducibility. They also confirm that views calibrated using the proposed method can be used to measure distances with millimetre precision using multiple views of a segment to be measured. The proposed display modes have also been approved by an expert surgeon who confirms their relevance. A full integration of all the 3D visualization system components has to be carried out in order to obtain a framework featuring the offline calibration method, the online update of all intrinsic and extrinsic endoscope parameters, the segmentation and registration of the preoperative model obtained by MRI and finally the fusion of endoscopic images with the a priori spine model. Once this framework is developed, it will be possible to validate it in a real surgical setting during a surgery performed on an animal model such as a miniature swine.