Construction d'un modele per-operatoire 3D du rachis pour la navigation en thoracoscopie.

摘要

The aim of this study is to propose, develop and validate the steps for the preparation of the 3D model for the navigation during minimally invasive surgery of the spine. The preparation of the model involves a step of automatic segmentation of intervertebral disk and a registration step to obtain the pre-operative 3D model of the spine into the intraoperative position.;To investigate the precision of the proposed segmentation process, a second study was conducted. It is important to choose the proper MRI to obtained satisfactory results of the automatic segmentation of the intervertebral disks. The measurement of the volumes showed that the proposed method systematically under estimated the size of the disks and that in the sagittal plane, the distance between the more posterior points of the segmented disk and to the one obtained from the manual segmentation, has a mean value of 1.8 mm (+/- 0.8 mm). Lastly, the results also showed that the inter users variability (manual segmentation) is slightly superior to the variability between the manual segmentation and the automatic segmentation. This confirms that the automatic segmentation is as reliable as the manual segmentation done by experts.;Following the automatic segmentation of the intervertebral disks, is a study on the 3D/2D registration between pre-operative MRI data of the patient lying supine and intraoperative X-ray data of the patient lying on the side. This study investigates the use of an articulated model in a registration context involving semi rigid structure like the spine. The validation of the 3D/2D registration technique is done in two phases. The first phase consists of using simulated intraoperative data with the addition of normally distributed random noises. The results of the simulations showed that the mTRE 3D error is of 4.42 mm (+/- 0.83) for a noise level corresponding to an error 0.9 mm in the manual identification of control points. The results of the second phase of the study showed that for the three clinical cases of scoliotic patients, the mTRE 3D is of 3.38 mm. Hence the study demonstrates that it is possible to achieve a 3D/2D registration with a single intraoperative X-ray of patient presenting a change of posture if a constraint is used in the optimization process.;Finally for a patient presenting a change of posture, the rigid registration of the control points located at the interface disk/vertebra has to be followed by an elastic registration so that the new space between two consecutive vertebrae can be filled up properly with the intervertebral disk. A preliminary study illustrated that it was possible to deform the disk with the same control points as the ones used for the rigid registration. The proposed technique consists of a 3D/3D elastic registration perform with kriging combined with a criteria of cross validation to automatically choose the optimal combination of polynomial able to best predict the type of deformation occurring in the different spine level during a specific change of posture. The preliminary study illustrates that from 10 control points and over, the results of the proposed elastic deformation method do not change significantly. Also, using three simulations of different deformations and 10 control points for each intervertebral disk, results showed that the choice of the type of polynomial used in the kriging equation is important since the mean error on the validation points can be up to three times higher depending on the type of polynomials used. (Abstract shortened by UMI.);The first part of the study proposed an automatic segmentation of intervertebral disks from scoliotic patient on MRI. This automatic segmentation based on the watershed technique and morphological operators is combined to a pattern recognition technique using texture features of the images. The clinical study done on a cohort of 11 scoliotic patients, showed that it was possible to automatically segment intervertebral disks and the over segmentation problem can be eliminate with the classification process.