Accuracy of registration of PET, SPECT and MR images of a brain phantom.

Accuracy of a surface-fitting algorithm for three-dimensional image registration of single photon emission computed tomography (SPECT), positron emission tomography (PET), and magnetic resonance (MR) images was tested using a three-dimensional, water-fillable brain phantom. Multislice or volume image sets were acquired for each modality. Small fiducial markers were attached to assess accuracy of surface fitting and provide an alternate fitting technique. A maximum gradient technique was found to work well for SPECT and PET edge detection. Transformation parameters for translation, rotation and scaling were determined by surface fit to match each SPECT and PET scan with MR images. Using the markers, overall translation errors were found to be < 2 mm in each direction and rotational errors < 2 degrees in every case. Errors for specific internal regions were also determined to be < 2 mm for most regions, with only a few fits resulting in errors > 3 mm for some cortical regions. Results indicate surface fitting to be sufficiently accurate for visual comparison of registered images and for enhanced SPECT and PET region of interest (ROI) determination and image reconstruction.