Content area
Abstract
This document proposes a methodology for generating a complete geometric model of the trunk of a patient with scoliosis. The geometric model is developed using the non-rigid registration of X-ray, TP and MR images, while taking into account the lack of anatomical correspondences between the image modalities, and the non-rigid deformation that occurs due to a posture change between the image acquisitions.
In the first phase, the shape of the spine extracted from MR images is registered to that extracted from the X-ray images all while compensating for spine shape changes that are due to the difference in posture between the acquisition of the two modalities. The semi-rigid transformation of the spine is obtained by means of an articulated model registration which is defined as follows: For each vertebra, a local coordinate system is constructed from vertebral landmarks. Local rigid inter-vertebral transformations are then obtained by computing the transformations between the local coordinate systems of adjacent vertebrae. Finally, the global transformation between each vertebra extracted from the MR images and the corresponding vertebra extracted from the X-ray images is obtained by concatenating the local transformations.
In the second phase, the TP, X-ray and MR images of the same patient are registered in order to obtain a full geometric model of the entire torso which incorporates the bone structures, soft tissue, as well as the external surface of the trunk. Firstly, the TP and X-ray images are aligned using a thin-plate spline and landmarks placed on the surface of the trunk of the patient prior to the acquisition of the two imaging modalities. Secondly, MR images are incorporated into the model using the articulated model followed by a thin-plate spline registration constrained in order to maintain the stiffness of the vertebrae. The quality of registration between the MR and the TP images is verified for 3 patients with scoliosis with the Dice index, which measures the overlap between the MRI slices and the space contained within the TP image. The Dice index varies between 0 and 1, where the value of 0 indicates that there is no overlap and a value of 1 indicates a perfect overlap. A value of 0.7 is considered suitable overlap. The proposed method is compared to rigid registration and registration a simple articulated model. An average Dice value of 0.95 is obtained when the proposed method is used, showing excellent overlap and a significant improvement compared to 0.82 in the case of simple articulated model registration and 0.84 in the case of rigid registration. Therefore, the proposed registration method succeeds in incorporating bone structures, soft tissues, and the external surface of the trunk using MR, X-ray and TP images all while compensating for the change in posture that occurs between these acquisitions.
In the third phase, inter-patient registration allows for the completion of a personalized three-dimensional partial model of the trunk of a patient by registering TP and X-ray images of the patient with the MR images of a generic model that is obtained by following the proposed methodology. Firstly, a patient having a full geometric model which incorporates the bone structures, soft tissues, as well as the external surface of the trunk is designated as the generic model. Secondly, a partial personalized model of another patient is obtained by registering the X-ray and TP images of the patient using a thin-plate spline function. Thirdly, MR images of the generic model are incorporated into the partial personalized model of the test patient using the articulated model transformation and the constrained thin-plate spline deformation. The Dice index is used in order to measure the overlap between the TP images of the patient and the MR images from the generic model following inter-patient registration.
The complete model obtained includes the bone structures, soft tissues and the complete surface of the trunk of scoliotic patients. This model can be incorporated into the surgical simulator which is under development, in order to take soft tissues into account while simulating the effect of spine instrumentation on the external surface of the patient’s trunk. However, the precision of the registration can be improved by using a 3 dimensional adaptive mesh of the soft tissues all while incorporating tissue-specific stiffness factors. (Abstract shortened by UMI.)
