CT/MR Synthesis

Synthesis of CT and MR Images of the Cervical Spine

Jeffrey G. Jarvik, MD, MPH

The magnetic resonance (MR) scan, shows neural elements (spinal cord and nerve roots) exquisitely. However, it is difficult to differentiate accurately between bone and soft tissue on MR images alone, particularly in the setting of degenerative changes such as dessicated discs, ossified ligaments, and arthritic joints. When surgery is contemplated, surgeons frequently obtain a cervical computed tomography-myelogram (CTM) to assist with planning. Because computed tomography (CT) delineates bony boundaries accurately, and because intrathecal injection of contrast (myelography) shows the extent of the dural sac and nerve root sheaths, CTM allows more precise assessment both of the severity of compression and of the nature (bone vs. soft tissue) of the tissues contributing to the pathologic changes.

Recently, powerful techniques for fusing cross-sectional images from different modalities have been developed. These techniques allow information gathered from different imaging modalities to be presented concurrently. For example, combining tissue density information obtained from CT scans with soft tissue information from MR scans could allow more precise image interpretation than either modality alone. The central hypothesis of this study is that a fused plain CT and MR image of the neck will provide information for diagnostic and surgical planning purposes equivalent to the MRI/CTM combination, while avoiding the expense and morbidity of intrathecal injection.

With this goal and hypothesis in mind, we have formulated two specific aims:

Specific Aim 1: Demonstrate the feasibility of CT-MR image registration in the cervical spine. The feasibility of precisely matching CT and MR image (registration) with subpixel accuracy has been demonstrated in the brain, which behaves as a rigid body to a very high degree of approximation. The registration transformation can therefore assumed to also be a rigid body transformation. The spine is only segmentally rigid, however, and so a more general class of transformations must be used for registration. We will develop the blended rigid body transformation, a mix of several rigid body transformations (one for each vertebra), blended smoothly in overlapping regions, for this purpose, and use the technique of mutual information maximization for the registration. Using techniques we have already used for CT-PET registration, we will estimate the accuracy of the registration. Our goal is a registration accuracy of 1 mm or less throughout the cervical spine, which we believe will be sufficient for Specific Aim 2.

Specific Aim 2: Show that fused CT-MR images provide presurgical planning information equivalent to that provided by CTM. We will select 50 consecutive patients who have undergone both MRI and CTM for degenerative disease of the cervical spine. CT images and MR images will be downloaded to a workstation and the CT images edited to an appearance comparable to that of ordinary (noncontrast) CT images. For each patient, a standardized clinical vignette will be developed and presented along with either CT-MR fused images or the MRI, CTM, and plain myelography images. Each of 7 experienced spine surgeons will be asked to formulate a detailed surgical plan based on the information presented. We will demonstrate equivalence of the surgical plans based on the fused images and those based on the MRI/CTM images. A similar study assessing diagnostic agreement will be performed with 4 neuroradiologists, who will also assess the accuracy of registration.