Results of a multi-institutional benchmark test for cranial CT/MR image registration

Abstract

Purpose: Variability in computed tomography/magnetic resonance imaging (CT/MR) cranial image registration was assessed using a benchmark case developed by the Quality Assurance Review Center to credential institutions for participation in Children's Oncology Group Protocol ACNS0221 for treatment of pediatric low-grade glioma.

Methods and materials: Two DICOM image sets, an MR and a CT of the same patient, were provided to each institution. A small target in the posterior occipital lobe was readily visible on two slices of the MR scan and not visible on the CT scan. Each institution registered the two scans using whatever software system and method it ordinarily uses for such a case. The target volume was then contoured on the two MR slices, and the coordinates of the center of the corresponding target in the CT coordinate system were reported. The average of all submissions was used to determine the true center of the target.

Results: Results are reported from 51 submissions representing 45 institutions and 11 software systems. The average error in the position of the center of the target was 1.8 mm (1 standard deviation = 2.2 mm). The least variation in position was in the lateral direction. Manual registration gave significantly better results than did automatic registration (p = 0.02).

Conclusion: When MR and CT scans of the head are registered with currently available software, there is inherent uncertainty of approximately 2 mm (1 standard deviation), which should be considered when defining planning target volumes and PRVs for organs at risk on registered image sets.

Figure 1

Transverse, coronal and sagittal slices of the CT and MR image sets used for the registrations.

Figure 2

MRI sections on which the target volume, the contrast enhancing lesion in the posterior occipital lobe, was to be outlined.

Figure 3

A sample registration result showing the target volume, as drawn on the MRI scan, projected onto the CT scan. The coordinates of the center of the target volume (in the CT coordinate system) were to be reported.

Figure 4

The coordinates in the CT coordinate system of the center of the large rod on the BRW frame (indicated by the arrow) were reported to define a common reference point.

Figure 5

Distribution of errors indicating the number of submissions falling within 5 mm error bins.

Figure 6

Distribution of results in the axial, coronal, and sagittal planes relative to the “true” center of the target.