Noninvasive Magnetic Resonance to Three-Dimensional Rotational X-Ray Registration of Vertebral Bodies for Image-Guided Spine Surgery

Everine B. van de Kraats, MSc; Theo van Walsum, PhD; Jorrit-Jan Verlaan, MD; F. Cumhur Öner, PhD; Max A. Viergever, PhD; Wiro J. Niessen, PhD


Spine. 2004;29(3) 

In This Article

Abstract and Introduction

Study Design: Magnetic resonance (MR) and three-dimensional rotational x-ray (3DRX) images of cadaveric spinal segments were registered using a conventional point-based technique and a noninvasive technique called maximization of mutual information (MMI).
Objectives: To assess the feasibility and accuracy of MMI-based registration in comparison with point-based registration as a new noninvasive image-to-patient registration technique for use in minimally invasive image-guided spine surgery.
Summary of Background Data: In image-guided orthopedic surgery, correspondence between before surgery acquired images and the patient is required. Currently, this necessitates an invasive registration step, in which anatomic landmarks on the bone surface have to be physically touched by a pointer. To overcome this invasive procedure, we propose using a calibrated 3DRX system, a modality that can visualize high contrast objects intraoperatively and that provides a direct correspondence between the image data and the patient. A noninvasive voxel-based technique is used to register the intraoperative 3DRX image to a before surgery acquired MR image.
Methods: Cadaveric thoracic and lumbar spine segments were implanted with markers, which were used as landmarks. The accuracy of the landmark-based registration was compared with MMI-based registration using the residual errors on the marker positions.
Results: The errors made using the point-based registration technique were compared with the errors made with MMI. The results show a statistically significant lower error (P < 0.01) for the proposed MMI method.
Conclusions: Noninvasive MMI registration of intraoperative 3DRX images to preoperative MR images is more accurate than point-based registration in cadaveric spine segments. It is therefore a promising technique for replacing the invasive landmark-based registration that is currently used in image-guided spine surgery.

Image-guided orthopedic surgery has the potential to improve accuracy, increase safety, and limit invasiveness of spinal procedures in which navigation is important. The use of image-guided surgery could reduce x-ray dose and improve three-dimensional (3D) insight during surgery. Image-guided surgery can be performed using preoperative magnetic resonance (MR) data, preoperative CT data,[1,2,3,4,5,6,7] and (multiple) tracked C-arm images.[8,9,10] A number of these studies have shown that the use of image guidance indeed improves accuracy[1,4,5,6,9] and increases safety.[5] Navigation on preoperative CT and/or MR data requires registration between the patient and the data volumes. In current image-guided surgery systems this registration process requires interaction by the surgeon, using either a point-based or a surface-based approach. In point-based registration algorithms, corresponding anatomic landmarks in image and patient are pinpointed. The patient is registered to the image by linking the landmarks. In surface-based registration algorithms, first a surface model of the bone in the preoperative image is made. Points that can be matched to this surface model are acquired by following the bone surface in the patient. Both methods are necessarily invasive: the anatomic landmarks or surfaces need to be exposed by removing soft tissue to exactly match the image to the patient. In some cases, this exposure goes beyond the dissection necessary for conventional surgical approaches. Given this disadvantage of these registration methods, a large interest exists for noninvasive registration techniques that could eventually lead to the development of less invasive or even percutaneous procedures. A possible method to noninvasively register the patient to the preoperative data is to acquire intraoperative two-dimensional (2D) fluoroscopic images that are subsequently coregistered with the preoperative 3D volume to obtain the orientation of the patient. These 2D-3D registration methods can be subdivided in feature-based[11,12,13,14] and intensity-based[15,16,17,18,19] methods. Feature-based methods extract features (e.g., bone edges) from the 3D image and 2D images, which are subsequently used for matching. Intensity-based methods use gray values from the fluoroscopic images and from projection images that have been simulated from the 3D volume.

This article proposes an alternative method to noninvasively register preoperative MR data to the patient. To this end, a calibrated 3D rotational x-ray system (3DRX) is used, which can visualize high contrast objects at high isotropic resolution.[20] The movement of the 3DRX system is calibrated and has a fixed geometry, which implies that all 3D images acquired with the system visualize the same physical space of the intervention room. If the position of the patient in the intervention room is known, e.g., by using an optical tracking system, a direct relation between 3DRX volume and patient will result. Subsequently, if the relation between MR data and 3DRX data is known, the relation between the MR data and the patient can be determined. Finally, this could allow navigation on both 3DRX and MR data, without requiring an invasive registration step.

In this work the accuracy of noninvasive MR data to 3DRX data registration is investigated. The accuracy is compared with marker-based registration. Marker-based registration is currently used for navigation during cranial interventions, and it is comparable to the landmark-based registration more frequently used in spinal surgery. If the accuracy of the noninvasive technique is at least similar to the marker-based technique, the proposed noninvasive registration technique could replace the conventional registration.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.
Post as: