Clinical Image Coregistration Variability on a Dedicated Radiosurgery Unit

On a new dedicated radiosurgery unit enabling frameless treatments, a cone-beam computed tomography (CBCT) can be used for stereotactic definition. Since magnetic resonance imaging (MRI) is used to delineate target, reproducible MRI-to-CBCT coregistration is vital for accurate target localization.

To evaluate reproducibility of image coregistration in patient images.

Three types of coregistration (source-to-target) were analyzed: (1) MRI-to-CT; (2) MRI-to-CBCT; and (3) CT-to-CBCT. For each patient (n = 15), each coregistration type was independently performed 5 to 30 times (total: 465 coregistrations). Each coregistration yielded a transformation matrix, which was subsequently applied to transform every point in the source image to stereotactic coordinates. Two metrics were measured: (1) target registration error (TRE): mean distance between the registered position of each target point and the average registration position of that point; (2) compound registration error (CRE): mean spatial difference between stereotactic coordinates using (A) MRI-to-CT-to-CBCT and (B) MRI-to-CBCT.

The median (range) of TRE was 0.11 mm (0.06-0.22 mm), 0.17 mm (0.10-0.36 mm), and 0.12 mm (0.08-0.21 mm) for MRI-to-CT, MRI-to-CBCT, and CT-to-CBCT, respectively. The TRE for MRI-to-CBCT was statistically higher than the other 2 methods (P < .01). The median (range) of CRE was 0.44 mm (0.22-0.59 mm). The maximum point CRE between patients ranged from 0.37-1.15 mm when considering all MRI points, but reduced to 0.31-0.90 mm within the central 16 cm. The CRE varied across the image volume, and typically was minimized near the center.

The variation in image coregistration is within 0.2 mm, indicating a high degree of reproducibility. The CRE varies throughout the head but is submillimeter in the central 16 cm region.