Image fusion is here. Practical, not terribly expensive, and rapidly disseminating in the imaging community are ways to volume register available DICOM-III standard images. Data from computed tomography (CT), magnetic resonance imaging (MRI), single-photon computed tomography (SPECT), and positron-emission tomography (PET) can be overlaid voxel by voxel with commercially available software programs on displays that allow the imaging physician to take advantage of the unique strengths of two entirely different modalities. Most fusion images now come from SPECT or PET superimposed on CT or MRI scans. This strategy combines tissue characterization or physiologic measurement from nuclear medicine with anatomic data from radiology. The radiopharmaceutical becomes the ultimate contrast agent. Fusion scans can also be easily formed from two nuclear medicine scans or from a CT and MRI scan, etc.

The fusion software typically calls for two sets of data from the same patient selected for fusion. One is selected as the "primary" set to which the other will be registered. It is best to keep the original image resolution of each image set and retain its quantitative information. The two blocks of data used do not start as registered images. The patient is scanned at different times, in different machines, and in different positions. Different size acquisitions with different slice thickness and pixel sizes with different central points are the norm. Voxel dimensions can be adjusted from measurements provided by calibration and quality control images. Adjustments required for registration take into account the X, Y, and Z coordinate offsets between the data blocks. The real trigonometry delight is correction of angular offsets between the two blocks of data. This kind of program is not a trivial software project (figure 1).

Not surprisingly, the alignment steps leading to coregistration can be guided effectively by the trained eye using interactive software. Some argue for the increased objectivity offered with a little forethought and the use of feducial markers detected by both modalities. A small point source of radioisotope mixed with Vitamin E works for SPECT/MR fusion scans. A point source of F-18 fluorodeoxyglucose (FDG) in a radiographically opaque cup works for PET/CT fusion. These feducial points placed on the surface of the patient are steered together with software that superimposes them to achieve accurate coregistration. Theoretically, only three feducials are needed to coregister two volumes of data. Many fusion imagers use a minimum of six. In reality, these surface feducials are a simple case of anatomic structures within the patient that are visible to each separate modality. For example, a pair of kidneys seen on an In-111 octreotide scan can be registered quickly by operator guidance with a CT or MRI scan that shows the kidneys as well. Matching a chest CT to a Tc-99m MAA perfusion SPECT is easy. Many scans have plenty of common structures, but there is a question of the operator's subjective interference with the process. He or she may want to "slide" an abnormality from a SPECT or PET scan on top of a particular anatomic finding on a MR or CT scan.

Automated software programs are in development that use information common to both scans. Mutual information algorithms have been developed that seek the minimum differences or maximum similarities between two cubic data sets from different modalities. Through an iterative process, these programs can bring data sets objectively into registration. The development of the transmission scan came from the need to measure attenuation in SPECT and PET scans. The transmission scan is coregistered with the emission scan automatically and is actually a low-resolution version of a CT scan. It has a lot of information in common with a CT scan. If a CT scan and transmission scan are fed to the right algorithm, they can be coregistered precisely without the operator's subjective opinion. The registration parameters derived for the transmission scan are applied to the emission scan to produce a fusion of the images (figure 2).

Display of fused images usually requires an overlay of a gray-scale image for the anatomic image (CT or MR) and a color scale for the nuclear medicine image (SPECT or PET). Dithering the two images so that every other pixel is displayed from each coregistered study on a 50-50 basis is the solution favored by the authors. Side-by-side comparison with interrogation of both coregistered images by regions of interest (ROIs) drawn on one or the other is helpful in some cases.

Nuclear medicine SPECT scans done with Ga-67, In-111 octreotide, I-131 and I-123, and Tc-99m agents (among others) are remarkably improved with the extra effort to perform fusion images (figures 3 through 7).

Almost every imaging modality has been used to evaluate pulmonary emboli. Fusion imaging may help address the problem of the "intermediate probability" Tc-99m MAA perfusion lung scan. CT of the chest can display the lung parenchyma to show areas of infiltrate, blebs, bullae, and tumors that cause defects on MAA scans that are not embolic.

Perfusion defects on the coregistered MAA SPECT scan that are explained by the CT/MAA fusion scan can logically be excused from the analysis of emboli. A prospective trial is needed to see how accurate this assumption is (figure 8).

Like their SPECT counterparts, PET scans are anatomy sparse images that can profit greatly from fusion imaging (figures 9 and 10). In the case
of targeting for purposes of radiotherapy, new developments allow PET/FDG images to be fused with CT radiotherapy simulations for planning treatment. This means that post-obstructive atelectasis behind a metabolically active (FDG-positive) tumor in the lung can be logically excluded from treatment. Radiation doses to the tumor can be increased, while doses to benign tissue can be decreased.

Using postacquisition volume registration software, Test/Retest paradigms from SPECT or PET studies may be done on different days and still compared voxel by voxel. Pre- and post-Acetazolamide (Bedford Laboratories, Bedford, OH) images from Tc-99m brain perfusion SPECT can be done in a semi-quantitative manner. The vasodilated Diamox SPECT is coregistered with the baseline study. The arithmetically different images are multiplied by 100, divided by the baseline images, and displayed as a percent perfusion reserve image. This image, in turn, can be displayed as an overlay on coregistered MR images.

MR images of primary bone tumors show exquisite detail of the soft tissue components of the tumor. CT images of the same bone tumors are much better at displaying the calcified cortical, trabecular, and tumor matrix details. A fusion of the CT bone windows with an MR provides optimum depiction of the principle elements of the tumor. This technique needs to be explored in depth.

Hybrid machines that acquire a PET and a CT scan, or a SPECT and a CT scan, are coming to market. These sequentially scan with two different modalities in the same gantry. One manu-

facturer uses a helical CT in the same gantry with a ring PET detector. Another uses a low output X-ray tube scanning in a circular track next to a pair of opposed nuclear medicine detectors for SPECT or coincidence detection. In each case, the patient is transported between these adjacent detectors on a moving tabletop. Attenuation correction for the nuclear medicine scan can be performed with the CT scan data. Speed of acquisition, examination inflexibility, resolution, and expense are limitations of these devices. Unless some of the laws of physics are suspended, there will not be a way to combine MR technology in the same gantry with CT or nuclear medicine. Fusion with MR requires separate acquisition of the other modality.

Even before formal studies are available in the literature, the fusion image that combines a nuclear medicine image and an anatomic image, or two nuclear medicine images, or two anatomic images, helps with the three "C's". Findings are made more conspicuous as one study (usually the nuclear medicine image) provides enhancement of the findings on the other. Findings are clarified as one study is used to explain the findings on the other (and this "C" works both ways). Diagnostic certainty improves, as the reading physician becomes more confident as a result of viewing the fusion images. Fusion imaging has arrived. AR