The registration of 3-dimensional (3-D) anatomical surfaces to sensor data such as intraoperative fluoroscopy is one of the basic problems in computer integrated surgery. The main objective is to find the relationship between 3-D preoperative computed tomographic images and a pair of intraoperative fluoroscopic images. Consequently, surgical navigation devices can use this relationship to provide improved surgical guidance. The proposed registration strategy presents a noninvasive anatomy-based (frameless) method for registration. In this article, we propose a cooperative approach between registration and contour segmentation on fluoroscopy. This approach is based on the duality between registration and segmentation in a model-based vision system. It associates a likelihood value to each pixel that corresponds to the probability that the pixel belongs to the contour of the object of interest. The registration is then achieved between backprojection lines stemming from likely contour pixels and the 3-D surface model of the object of interest. Then, in order to take into account the internal contour points extracted by the cooperative approach, we propose a new line to surface distance computation algorithm to be used during the data to model distance minimization step. Finally, we present the obtained results that demonstrate the validity of the proposed approach in carrying out accurate 3-D and 2-D registration.

Several attempts have been made to determine the motion of the lumbar spine with the purpose of gathering additional data for the diagnosis of instability. A pair of functional radiographies taken at each end of the range of motion can be used to carry out the estimation of motion using graphic construction or computer assisted methods. In this paper we propose to develop a new method to measure the motion of the lumbar vertebrae using 3D/2D registration between 3D surface models and a pair of calibrated 2D X-ray images.

When inserting screws into a vertebral pedicle, the surgeon usually exposes the back part of the vertebra and uses his or her anatomic knowledge to align the drill in the proper direction. A slight error in direction may result in an important error in the position of the tip of the screw. This is done with no direct visibility of crucial structures (spinal cord, pleura, vessels). Statistical analysis of a series of surgical procedures has shown that 10% to 40% of the screws are not installed correctly. To reduce the risk of complication, a computer assisted method is proposed that enables the surgeon to place a screw at a position preoperatively defined in 3 dimensions using computed tomography images. This allows the surgeon to align a standard surgical drill with the optimal position and direction. The depth of the pilot hole during drilling also is monitored by the system to prevent penetration of the anterior cortex of the vertebral body. Using this procedure, in vitro tests were performed and showed that an accuracy of less than 1 mm can be obtained. Clinical trials were done in 10 patients who suffered severe scoliosis or spondylolisthesis. The trajectory of the holes drilled in L2, L3, L4, and L5 vertebrae were checked for all clinical tests. Postoperative radiographs and computed tomography scans showed that the screws were well inserted in each plane for each pedicle. This technique also can be used to perform osteosynthesis at the thoracic and cervical levels.

Passive and semi-active aids ince 1985, a team of computer scien-surgeons has been involved in a project at Grenoble Hospital called computer assisted medical interventions (CAMI). The aim of the project is to help surgeons and physicians use multimodal data in a rational and quantitative way in order to plan and to perform medical interventions. Recent advances in medical imaging systems such as CT and MRI have stimulated research on the interpretation of medical images. Nevertheless, very few systems allow for an efficient therapeutic use of the wealth of information these images contain , which is CAMI's twofold objective: =Define an operative strategy that takes advantage of the localizing capabilities of imaging, and make this strategy available in an operative reference system. This goal requires models and processing of basic data, in conjunction with a priori knowledge , in order to define an optimal strategy. =Perform the previously defined operative strategy, with the aid of a suitable guidance system, under appropriate imaging supervision. Without help, executing a strategy can sometimes be very difficult. First, reproducing a defined strategy directly raises problems, for one has to mentally match geometrical information observed in different reference systems (mainly intra-operative scenes or images with pre-operative images). Then, complex interventions may be necessary on organs difficult to reach or for which a human operator will only have poor or no visibility. In addition, surgical tools, such as probes, most often have to be very precisely positioned, and submillimetric accuracy may be required for microsurgery. Finally, some interventions may be dangerous for the medical staff (e.g., contamination or irradiation). In all these instances, optical or mechanical guidance systems are required. The ultimate success of this research hinges on complex robotics systems and their various sensors. The aforementioned objectives aim at improving the quality of the interventions by making it easier, more accurate, closer to a a pre-operative simulation where accurate objectives can be defined, and sometimes faster. Also, it may be possible to devise new interventions and to validate protocols of therapeutic research. Obviously , this is long term research, with many potential clinical applications. Yet, a general methodology can be applied to various clinical situations [ 11, as presented below. In order to understand the methodology and applications of CAMI, it is necessary to be aware of current technology. Many issues in the CAMI project are concerned with geometrical localization problems. Hardware issues mainly refer …

In the field of Augmented Reality in Surgery, building a hybrid patient's model, i.e. merging all the data and systems available for a given application, is a difficult but crucial technical problem. The purpose is to merge all the data that consitute the patient model with the reality of the surgery, i.e. the surgical tools and feedback devices. In this paper, we first develop this concept, we show that this construction comes to a problem of registration between various sensor data, and we detail a general framework of registration The state of the art in this domain is presented. Finally, we show results that we have obtained using a method which is based on the use of anatomical reference surfaces. We show that in many clinical cases, registration is only possible through the use of internal patient structures.

In Computed-integrated Surgery (CIS), the registration between pre- or intra-operative images, anatomical models and guiding systems such as robots or passive systems is a crucial step. In our methodology, rigid or elastic transformations are estimated using non-linear least-squares minimization of euclidean distances computed on data that can be 3D surfaces or 2D projections. This paper shows the variety of results that is achieved with this framework on several clinical applications.