5.3.2 Real Scans

The Hoffman brain phantom consists of 19 slices ( mm thick). Each slice additionally consists of 6 sub-slices which are cut out and stacked so that filling the cylindrical phantom containing all 19 slices with a bath of activity simulates the gray to white matter uptake ratio of cerebral metabolism [Dat]. The amount of plastic in a 6.4 mm thick slice is inversely proportional to the amount of blood uptake or activity. In other words, open space where the radioactive bath may permeate corresponds to the uptake of activity in the SPECT scan. In the MR scan, the signal only comes from the decayed radioactive bath so that the resulting image corresponds mostly to gray matter. The acquisition times and activity used for the SPECT scans were such that the total counts per projection (64 total projections) were typical of clinical scans of the brain (about 60 kilocounts/view).

Six external fiducials were symmetrically placed on the exterior of the phantom (see figure 5.1) so that their centroid was at about the geometrical center of the Hoffman brain (see MR and SPECT images in figure 5.2) which anatomically corresponds to the general location of the body of the corpus callosum. This allowed the specification of the rotational errors about the same centroid. The symmetry of the configurations is thus preserved more closely. This is extremely important because it allows direct comparison of the measured errors here to the results of the point simulation investigations. The fiducials were hollow latex ``voids'' with flared bases to allow attachment to the phantom surface with tape (see figure 5.3). The roughly spherical voids of the fiducials were about 5 mm in diameter. They were filled with a syringe containing a 10 milli-molar solution of CuSO in water and Tc-pertechnetate ( in each) while simultaneously vented with another syringe to allow the release of pressure. The fiducials were placed at the edges of the top and bottom plate (see figure 5.1) away from the activity which was about eight times larger than would localise in the brain in clinical scans ( at the time of acquisition for the simulation) so that streaking artifacts could be avoided in the resultant reconstructed images. The phantom was pre-filled 24 hours before scan time with a well mixed solution of about 300 mCi of Tc-pertechnetate in distilled water to ensure the absence of photopenic artifacts. The fiducial positions were simply discretised visually because other algorithms in the literature [TJG+91] didn't provide any significant advantages. The precision of the identification of the external marker positions using the mean of Gaussian fits, the maxima in in second order fits and visual identification are all no better than of the pixel dimensions. The accuracy of the position location is not of interest. The errors resulting from the inexactness of the marker location are inconsequential because it is the distribution of the errors which will be measured and not their mean. Exactly correct external marker locations would result in the mean of the distribution of registration errors being equal to 0, but the measure of the standard deviation as a description of the distribution of the registration errors is of more interest and is independent of the mean of the distribution.

The MR volume was acquired on a Phillips Gyroscan S15 1.5 Tesla scanner using a multi-slice spin echo sequence with an echo time of 20 msec and a repetition time of 2000 msec. It was reconstructed as a 25625664 image matrix with voxel dimensions of approximately 113 mm. The SPECT volume was acquired with a single-headed square crystal GE camera with a low energy high resolution parallel hole collimator and reconstructed as a 12812864 image matrix with voxel dimensions of about 2.72.72.7 mm. A Butterworth pre-filter was used with a power factor of 10 and cut-off at 0.48 cycles/mm. The two volumes were standardised to the MINC file format and display convention (see appendix) [Nee93] to allow simultaneous display of the three cardinal planes of each volume for interactive homologous point pair selection [MCEP89]. The anisotropic nature of the resolution of the SPECT system used is given in the table 5.1. These full width at half maxima measurements in air for a circular radius of rotation of cm were performed according to the accepted standards in the literature [SBJ+87] and calculated as Simplex fits to a Gaussian plus pedestal [Gro81]. The resolution measurements were made in air because there is no clear standard in the dimension of the phantom used for in medium measurements. The measurements in air are easily reproducible for comparison between institutions because they are independent of phantom dimensions. In phantom resolution measurements are also performed in medium with cold rods and spheres in uniform baths of radioactivity to avoid count problems, but those measurements do not yield full width at half maximum measurements which may be used to compare directly registration error descriptions which are also in terms of full width at half maximum calculations. This radial dependence is largely due to the decrease in the detection of large angle scatter as the source moves closer to the plane of the detector face. This ultimately gives rise to a radial dependence of the SPECT's resolution.

A disperse distribution of 25 homologous points were selected in both volumes. Similar to what was done in the point simulations, the assumption is made that the target volume (MR) which produced points set A has because of MR's superior resolution. This led to the approximation that was made in equation 4.4. This is a reasonable assumption because this is the very basis for the need for anatomical/functional registration. The same configuration of SPECT homologous points were chosen 25 times to produce registrations. As with the point simulations, and () practical registration transformations were obtained and subtracted from the same and provided by the external fiducial test points. and were then similarly found over and to characterise the registration errors for a clinically realistic as provided by the complex nature of the brain phantom in real scans.