a. Rectilinear scanner with focused collimators.

b. Anger tomoscanner (two opposed small cameras, converging collimators, raster scanning, multiple coronal whole-body tomographic images).

c. Anger camera with a seven-pinhole collimator (cardiac short-axis images).

d. Planar gamma camera imaging is weakly tomographic.

a. Planar gamma camera images are acquired over 180° (cardiac) or 360° around the patient.

b. Transverse images are formed by filtered backprojection (FBP) or iterative reconstruction.

a. Continuous mode: images acquired over angular intervals as the head moves continuously

b. “Step-and-shoot” mode: images acquired when the head stops at evenly spaced angles

c. Pixel format usually 64² or 128², but large enough to not degrade resolution due to pixel size

d. Zoom factor typically approximately 1.5 for 64² cardiac, to reduce pixel size and, thus, improve resolution

e. Typically, 60-64 (64²) or 120-128 (128²) projections, to reduce reconstruction streak artifacts

f. Circular orbit: brain imaging (on systems with a small camera head physical edge-to-useful field-of-view edge distance or on systems with a head holder beyond the patient table; resulting in higher resolution compared to body imaging, due to a small radius within the patient’s shoulders)

g. Noncircular (“body contouring”) orbit (NCO): body imaging (for a circular orbit, camera heads would be many centimeter from the body surface for anterior and posterior projections, resulting in a loss of spatial resolution). NCO manually specified on some older systems, but automatically determined on modern systems.

a. Projection images are mathematically filtered.

b. Each transverse slice is formed by backprojection of the corresponding row of projections.

c. The theoretical ideal kernel is a ramp filter in the spatial frequency domain.

d. Projections contain substantial statistical noise (radioactive decay process, low pixel counts).

e. High frequencies in the spatial frequency domain are dominated by statistical noise.

f. Gamma camera resolution reduces higher spatial frequency information of imaged structures, increasingly with distance.

g. Ideal ramp FBP results in unacceptable noise in transverse slices.

h. Projections are smoothed by ramp filter “roll-off” at higher spatial frequencies.

i. Filter kernel “roll-off” reduces noise at the expense of spatial resolution.

j. Filter kernel selection is a compromise between resolution and noise.

k. Different kernels are employed for different SPECT studies, based on the spatial resolution (due to collimator and detector-to-imaged organ distance) and amount of statistical noise (due to amount of activity injected, collimator and acquisition time per projection).

l. A filter with a higher spatial frequency cutoff is required for projections of better spatial resolution and less quantum mottle, to avoid unnecessary reconstruction resolution loss.

m. A filter with a lower spatial frequency cutoff is required for projections of poorer spatial resolution and greater quantum mottle, to avoid excessive reconstructed quantum mottle.

a. Filtered backprojection is computationally efficient but assumes perfect projection images of a three-dimensional object, which is not true, due to photon attenuation by the patient, the inclusion of Compton scatter, and collimator resolution degradation with distance.

b. The initial activity distribution assumption is uniform or the result of simple backprojection.

c. Projection images of the initial distribution are then calculated and compared to the actual (acquired) projection images.

d. The activity distribution is adjusted based upon the projection image comparison.

e. Distribution projection, projection comparison, and distribution adjustment are repeated until the calculated projection images are approximately equal to the actual projection images.

f. Projection calculation can incorporate and thus compensate for

g. Reconstructed images can be of either higher quality than FBP images or equivalent quality with less administered activity or reduced scan time.

h. Iterative reconstruction is computationally less efficient than FBP, but feasible due to increasing computer speed, the small SPECT matrix sizes, and computationally efficient algorithms such as ordered subset expectation maximization.

i. Three-dimensional spatial filtering is usually applied afterward, due to substantial noise in images reconstructed from projections with relatively poor counting statistics.

a. The Chang approximation method for FBP assumes a constant µ value; may be used for brain and abdominal studies where µ is relatively constant; cannot attenuation correct thorax studies, where µ is highly variable; and should be verified (including µ value) with a phantom before clinical use (Fig. 19-5, middle and right).

b. Sealed radioactive sources (commonly Gd-153, which emits 97- and 103-keV γ-rays) to measure attenuation were added by several SPECT camera manufacturers in the 1990s:

c. Hybrid SPECT/CT scanners, where x-ray CT image data can be used for radionuclide emission data AC, have supplanted radioactive transmission sources.

d. Extensive studies of radioactive source- and x-ray CT-based myocardial perfusion SPECT AC have demonstrated a reduction in attenuation artifacts and improved diagnostic performance.

e. Other studies have shown that emission data and µ map spatial misalignment can cause AC artifacts in myocardial perfusion SPECT images.

f. A period of transition is needed to retrain even experienced clinicians to interpret attenuation-corrected myocardial perfusion SPECT images.

a. Coronal and sagittal slices are commonly generated by simple reordering of the pixels from the transverse slices.

b. Oblique images parallel (vertical and horizontal long-axis images) and perpendicular (short-axis images) to the long axis of the left ventricle are standard for cardiac SPECT.

c. Due to considerable anatomic variation among patients, the long axis of the heart must be determined before the computer can create the oblique images.

d. The long axis is commonly defined manually by a technologist, although the software on most systems is now capable of correct automatic reorientation, with operator verification and override.

a. Double-head, 180° opposed heads: head and body SPECT, whole-body (WB) planar

b. Triple-head, fixed-angle: head and body SPECT, but limited crystal width (no WB planar)

c. Double-head, variable-angle: head and body SPECT and WB planar (180° opposed), and cardiac SPECT (90° perpendicular)

a. Veriton general purpose (Spectrum Dynamics Medical, Fig. 19-7 left)

b. NM530c dedicated cardiac (GE Healthcare, Fig. 19-7 middle)

c. D-SPECT Cardio dedicated cardiac (Spectrum Dynamics Medical, Fig. 19-7 right)

d. CardiArc dedicated cardiac (CardiArc Inc.)

a. A 22-cm diameter, cylindrical plastic water-filled phantom.

b. Three Co-57 solid or Tc-99m liquid line inserts (one on the central axis, two offset by 7.5 cm).

c. A 15-cm radius circular orbit, 360° acquisition; and ramp FBP (a “roll-off” reduces resolution).

d. Typical reconstructed FWHMs (LEHR collimator) are 9.5 to 12 mm (central), 7 to 8 mm (tangential), and 9.4 to 12 mm (radial).

a. X and Y magnification factors (gains) relate distances in the object to numbers of pixels in the acquired image, determined by imaging two point sources a known distance apart, along a line parallel to the camera’s X and Y axes:

and should be equal to one another, as well as between detectors of a multihead system, to prevent projection, coronal, sagittal, and oblique image distortion.

b. Multienergy spatial registration is important in SPECT for both imaging radionuclides which emit photons of more than one energy (e.g., In-111, Ga-67), and for uniformity and axis-of-rotation (AOR) corrections determined with one radionuclide to be valid for others. Gain and multienergy registration errors are not applicable to multielement cameras, as detector elements are square and the number of element rows and columns is fixed.

c. Center-of-rotation (COR) calibration aligns projection images to the AOR (an imaginary line about which the SPECT camera revolves).

d. Uniformity of a camera head is important, and more so for SPECT versus planar imaging.

e. Camera head tilt causes a loss of reconstructed spatial resolution and contrast (less toward the center and greatest toward the edge), as out-of-slice activity will be backprojected into each transverse slice (Fig. 19-10), as the camera head is assumed to be aligned with the AOR (exactly parallel for most collimators).

a. Advantages:

b. Disadvantages:

SPECT/CT is supplanting SPECT-only imaging, as it has become mainstream and systems have proliferated. SPECT/CT does require the additional QC step of SPECT and CT alignment verification in every clinical examination for optimum visual and quantitative accuracy.

a. Annihilation Coincidence Detection: Positron emission results in two 511-keV photons that are emitted in nearly opposite directions (Fig. 19-12). If both photons from an annihilation are detected in the scanner at nearly the same time, a process called ACD, then the line in space connecting the locations of the two interactions, which is known as a line of response (LOR), establishes the trajectories of the detected photons.

b. True, random, and scatter coincidences: Figure 19-13

c. Detection of interactions: Scintillation crystals optically coupled to photomultiplier tubes (PMTs) or other light detectors are used. The signals from the PMTs or other light detectors are processed using pulse mode to create signals identifying the position in the detector, deposited energy, and time of each interaction. The energy signal is used for energy discrimination to reduce mispositioned events due to scattering and the time signal is used for coincidence detection. Original PET detector designs had one scintillator coupled to one PMT. However, due to cost and limitations on PMT size, current systems use a block of scintillation crystals coupled to a smaller number of PMTs (Fig. 19-14). The relative magnitudes of the signals from the PMTs coupled to the detector block are then used to identify the location of the crystal within the detector block that had a photon interaction, as in a scintillation camera. The ideal scintillation detector characteristics include prompt light output (short decay constant), high linear attenuation coefficient (stopping power), high conversion efficiency, and high energy resolution to discriminate against scattered events. Originally, BGO used to be the dominant scintillation crystal material in PET scanners. Recently, LSO (LYSO) has become the dominant crystal material.

d. Timing of interactions and detection of coincidences: The timing of the interaction of an annihilation photon with the scintillation detector is determined from the leading edge of the electrical signal from the PMTs. When the time signals from two detectors occur within a selected time interval called the time window, a coincidence is recorded. A typical time window for a system with BGO detectors is 12 ns. A typical time window for a system with LSO detectors, which emit light more promptly, is 4.5 ns.

e. True versus random coincidences: Randoms rate is defined by: R_random = τS₁S₂, where τ is the coincidence time window and S1 and S2 are the actual count rates of the detectors, often called singles rates. Randoms decrease linearly with decreasing timing window and increase proportionally with the square of the activity in the patient. This is contrasted with an approximate proportional increase in the true coincident rate.

f. Scatter coincidences: Scatter coincidences are dependent upon the amount of scattering material and thus occur less in head than in body imaging. Scatter coincidences in PET imaging can be very high (60% to 70% of the acquired coincidence data), particularly in larger cross-sections of the body and when data are acquired in three dimensions. The use of narrower energy discrimination window can reduce the number of scattered events but will also negatively affect the number of recorded true coincidences.

g. Two-dimensional and three-Dimensional data acquisition: PET coincidence data can be acquired in two-dimensional (slice) or three-dimensional (volume) mode (Fig. 19-15). In slice mode, coincidences are detected and recorded only within each ring of detector elements or within a few adjacent rings of detector elements. Septa (made from tungsten) prevents coincidences outside an axial slice from reaching the detector ring of that slice hence limiting the acquisition to two-dimensional mode. In volume mode, the absence of the septa allows coincidences to be recorded from the whole detector volume. In volume mode, the total number of recorded events is substantially increased (increase in scanner detection efficiency), but so is the number of recorded scattered and random events. In volume mode, the scanner sensitivity increases linearly from the edges of the scanner to the center of the scanner (Fig. 19-16). This variation in scanner sensitivity requires overlapping of bed positions in situations where patient imaging extends beyond the physical axial length of the scanner. Overlapping of bed position reduces the effect of this variability or the resultant images. Some PET systems are capable of acquiring data in two or three-dimensional mode using retractable septa, but most new PET systems can only acquire data in volume mode.

h. Transverse Image reconstruction: Following the acquisition of the LORs, the data are corrected for several confounding factors and reconstructed using iterative or analytical techniques. Image reconstruction in twodimensional mode follows similar approaches to SPECT, while three-dimensional mode requires techniques that capture the volumetric data.

a. Normalization correction: conducted to adjust for performance variability among the thousands of detectors in the scanner.

b. Randoms correction: conducted to subtract random events from the recorded total number of events. Two approaches are used: randoms assessment from delays or randoms assessment from singles with the latter resulting in a more accurate correction.

c. Scatter correction: conducted to subtract scatter events from the recorded total number of events. Two approaches are used: scatter is estimated from reconstructed PET and attenuation images while employing computer modeling of photon interactions or scatter is estimated from fitting smoothly varying continuous functions to the tails of the projection profile outside the imaged object. The first approach is mainly used on new hybrid PET/CT systems where the CT scan is used as the attenuation map.

d. Deadtime correction: conducted to account for pulse pileup and deadtime effects whereby various factors are applied to correct any deviation of the scanner count rate performance from a linear response with increasing activity concentrations.

e. Decay correction: conducted to correct for radioactivity decay during the imaging time (Eq. 19-2).