Given a SPECT and an MR brain volume data set of the same patient which requires registration by the ILM technique, the first task at hand is the identification of homologous points in each volume. The data format of the volumes produced by different scanners and different vendors must be standardised to allow a common facility of accessing both the raw data and all ancillary header information before this may be done. At the Montreal Neurological Institute (MNI), the identification of the homologous points is accomplished by using an application [MCEP89] which displays the three cardinal planes of each data volume side-by-side and manually ``tagging'' homologous points by panning through the two volumes in real time. Since, in practice, the homology of the points chosen for registration using the Procrustes algorithm are almost entirely limited by the quality of the lower resolution SPECT images, a brief overview of brain imaging with SPECT is given. SPECT images are reconstructed from parallel or fan-beam projections according to classical computed tomography principles which were first applied in nuclear medecine. There are many factors which affect the spatial resolution of the resultant tomographic images which will in turn affect the homology error between the points sets and, ultimately, the accuracy of the registration. These will be discussed in the context of the theory of linear systems. The homology error is also affected by the difference in the type of information the images from each modality provide and the ability of the operator choosing homologous point pairs to reconcile the two types of information. SPECT volumes physically correspond to a 3-D spatial map integral counting measurements of the radioactivity distributed in a subject's brain. In the brain, the spatial distribution of radioactivity corresponds to the areas of deposition by the blood and, because of the ligand bound to the radioisotope used, retention by the brain so that the SPECT images are representative of cerebral metabolism. MR volumes correspond to spatial maps of not only proton density but also to the rotational freedom of molecules which contain hydrogen as well as their proportion in the different body-fluid compartments. Point sets chosen from low and high resolution MR volume data sets will have smaller resultant homology errors because anatomy is being matched to anatomy whereas SPECT/MR registration requires the correspondence of points picked in metabolic and anatomical images (see figure 2.1). The primary tracers used in brain SPECT, the nature of the information provided, and some of their clinical applications are thus also discussed to better understand the nature of the information provided by SPECT images.
