Feasibility of using ultra-high field (7 T) MRI for clinical surgical targeting

Abstract

The advantages of ultra-high magnetic field (7 Tesla) MRI for basic science research and neuroscience applications have proven invaluable. Structural and functional MR images of the human brain acquired at 7 T exhibit rich information content with potential utility for clinical applications. However, (1) substantial increases in susceptibility artifacts, and (2) geometrical distortions at 7 T would be detrimental for stereotactic surgeries such as deep brain stimulation (DBS), which typically use 1.5 T images for surgical planning. Here, we explore whether these issues can be addressed, making feasible the use of 7 T MRI to guide surgical planning. Twelve patients with Parkinson's disease, candidates for DBS, were scanned on a standard clinical 1.5 T MRI and a 7 T MRI scanner. Qualitative and quantitative assessments of global and regional distortion were evaluated based on anatomical landmarks and transformation matrix values. Our analyses show that distances between identical landmarks on 1.5 T vs. 7 T, in the mid-brain region, were less than one voxel, indicating a successful co-registration between the 1.5 T and 7 T images under these specific imaging parameter sets. On regional analysis, the central part of the brain showed minimal distortion, while inferior and frontal areas exhibited larger distortion due to proximity to air-filled cavities. We conclude that 7 T MR images of the central brain regions have comparable distortions to that observed on a 1.5 T MRI, and that clinical applications targeting structures such as the STN, are feasible with information-rich 7 T imaging.

Conflict of interest statement

Competing Interests: Essa Yacoub is currently a PLoS ONE Editorial Board member. This does not alter the authors' adherence to all the PLoS ONE policies on sharing data and materials.

Figures

Figure 1. Workflow of data processing and analysis.

The workflow process of image post-processing and quantitative estimation of the global degree of correspondence between co-registered 7 T and 1.5 T MR images, using anatomical landmarks method.

Figure 2. Visual example of the registration results.

Edges of brain structures obtained from the registered 7 T image (right column) are superimposed on the 1.5 T image (left column). Top row: T1W coronal images acquired at (a) 1.5 T and (b) 7 T. Bottom row: T2W axial images acquired at (c) 1.5 T and (d) 7 T, respectively. Note the high degree of correspondence between the 7 T red iso-contour edge lines superimposed on the 1.5 T images, indicating minimal distortion in the 7 T images compared to the clinical 1.5 T images. This method follows the current standard practice used in the operating room, in which the surgeon overlays the registered image on top of the reference image and toggles between them in order to determine, visually, how well regional brain structures coincide in the different image sets.

Figure 3. Regional contribution to global registration.

a) Nine sub-volumes to which the T1W brain images were parceled for the local registration method. The regions are: 1. Posterior-superior, 2. Mid-superior, 3. Antero-superior, 4. Posterior-middle, 5. Middle (includes the midbrain and portions of the temporal lobes), 6. Antero-middle, 7. Posterior-inferior (corresponds mostly to the posterior fossa contents i.e., cerebellum), 8. Mid-inferior (corresponds primarily to the inferior portion and floor of the middle fossa) and 9. Antero-inferior (corresponds primarily to the anterior skull base and temporal poles). T2W images were parceled into three sub-volumes, which approximately correspond to the three middle regions of the T1W images (regions 4, 5 and 6). b) Average local skews of the parceled regions, superimposed on a surface rendering of a representative brain. The colors (cold-to-hot) reflect the amount of average regional skew required for a registration at that region (measured as the tangent of the skew angle). Note that distortions at the middle region, which include the midbrain and portions of the temporal lobes, are minimal and indicate the clinical applicability of 7 T imaging of these sub-volumes for DBS procedures for example. The registrations of the inferior regions (regions 7, 8 and 9) were typically unsuccessful due to loss of signal in these areas in the 7 T images, which did not allow meaningful estimation of distortion when using this method.

Figure 4. Quantitative and statistical summary of anatomical based registration.

Box plots of the statistics of the co-registered landmarks' distances, obtained from individual patients, for T1W (top row – a and b) and T2W (low row – c and d) images. Further comparison is done between images acquired by 7T/PS (left column – a and c) and 7T/AS (right column – b and d). The red lines represent the average, top and bottom of the box represent the first and third quintile, respectively, the whiskers show the maximum and minimum values, and the asterix represent the outliers. The statistics is of distances between the anatomical landmarks assigned on 1.5 T images and the corresponding anatomical landmarks assigned on the 7 T images, after a global landmarks-free coordinates transformation to the 1.5 T image space. As can be seen, the distances are on the order of one voxel, indicating a high degree of correspondence between the two co-registered images.

Figure 5. Global registration matrices values.

Transformations values for the registration used for the T1W (top row – a and b) and T2W (low row – c and d) and for 7T/PS (left column – a and c) and 7T/AS (right column – b and d) images, as derived from the corresponding transformation matrices. The transformation values are measured as maximum scaling change (deviation of the scaling form 1), maximum skew (measured as the tangent of the skew angle), and volume change (deviation of the transformation determinant form 1) that were needed for global registration between 7 T and 1.5 T images. The values indicate that the registration operators are essentially rigid body transformations with negligible linear corrections in scaling, skew, and the transformation determinant. Given the high degree of correspondence between the registered images, these values indicate minimal amount of geometric distortion between the 1.5 T and 7 T images. Greater amount of scaling and skew was recorded for the 7T/AS T2-weighted images (5d). This is due to a stronger gradient employed in the 7T/AS MR system.

Figure 6. Regional registration matrices values.

Average values (per brain region) of maximum scaling change (measured as the scaling deviation form 1), maximum skew (measured as the tangent of the skew angle) and volume change (measured as the deviation of the transformation determinant form 1), that were needed for regional registration between 7 T and 1.5 T images. Presented are the values of T1W (top row – a and b), T2W (low row – c and d), 7T/PS (left column – a and c) and 7T/AS (right column – b and d). These values represent first order approximations of the amount of distortion associated within each part of the imaging volume. As expected, the Posterior-superior, Antero-superior and Antero-middle, exhibit greater distortion, while central and mid-back regions presented minimal levels of distortion.

Figure 7. Registration of 7T MRI to CT.

An example of co-registration between CT and a) 1.5 T T1W, b) 7 T T1W, c) 7 T T2W. Although CT images are far less informative than MRI, it can be seen that the ventricles are perfectly aligned, indicating an adequate quality of registration. Registering 1.5 T MR images to a CT is a common practice for planning of neurosurgical procedures including DBS surgery and tumor resections. This practice allows for using the superior contrast of MRI while capitalizing on the geometric integrity of the CT image. Here we suggest that the same practice may be used for correcting for whatever geometrical distortions that may be present in 7 T MRI.