Comparison of manual and automatic MR-CT registration for radiotherapy of prostate cancer

Abstract

In image-guided radiotherapy (IGRT) of prostate cancer, delineation of the clini-cal target volume (CTV) often relies on magnetic resonance (MR) because of its good soft-tissue visualization. Registration of MR and computed tomography (CT) is required in order to add this accurate delineation to the dose planning CT. An automatic approach for local MR-CT registration of the prostate has previously been developed using a voxel property-based registration as an alternative to a manual landmark-based registration. The aim of this study is to compare the two registration approaches and to investigate the clinical potential for replacing the manual registration with the automatic registration. Registrations and analysis were performed for 30 prostate cancer patients treated with IGRT using a Ni-Ti prostate stent as a fiducial marker. The comparison included computing translational and rotational differences between the approaches, visual inspection, and computing the overlap of the CTV. The computed mean translational difference was 1.65, 1.60, and 1.80mm and the computed mean rotational difference was 1.51°, 3.93°, and 2.09° in the superior/inferior, anterior/posterior, and medial/lateral direction, respectively. The sensitivity of overlap was 87%. The results demonstrate that the automatic registration approach performs registrations comparable to the manual registration.

Figure 1

Sagittal view of the transformed MR, original CT, checkerboard representation of the registration result of the manual registration, and checkerboard representation of the registration result of the automatic registration for two patients.

Figure 2

The two contours are obtained from a bone and prostate stent segmentation in CT followed by an inverse transformation of the registration by the automatic (white) and the manual (black) registration for one patient. The segmentations are superimposed on the original ${\mathrm{T}}_2-\text{weighted}$ MR images: (a) bones and prostate stent segmentation in the axial plane; (b) region of interest and the prostate stent in the axial plane; (c) region of interest and the prostate stent in the sagittal plane. It can be noticed, in particular in image (a), that both methods are based on a registration of the prostate stent and a tightly surrounding volume as either the black or white contour follow the contour of the pelvic bone. This results in an accurate alignment of the prostate stent in image (b), which indicates that the prostate has moved relatively to the pelvic bones. In image (c), a gap in the segmentation of the prostate stent can be observed from the automatic registration caused by interpolation.

Figure 3

The order is sagittal, coronal, and axial views of the overlap of the manual prostate delineation performed in the original MR image and then aligned to the CT based on the manual registration (black), the automatic registration of the CTV (blue), and the automatic registration of the extended CTV (white). The delineations are superimposed on the CT images.