Digirad Cardius 3 XPO

The first of these systems to be introduced was the Cardius XPO line manufactured by Digirad Corporation (Poway, CA). Although the initial design of a similar Digirad system was described in 1998 with the use of cadmium zinc telluride (CdZnTe, or CZT),⁶ all manufactured systems use solid-state cesium iodide (CsI)-photodiode detectors. This system can be configured in 1-, 2-, or 3-detector configurations.⁷ The Cardius 3 XPO is shown in Figure 9-1. Current models make use of pixilated CsI(Tl) detectors and photodiodes to configure detector heads that are more compact than conventional cameras equipped with photomultipliers. Each detector head is 21.2 × 15.8 cm and contains an array of 768 6.1 × 6.1 × 5 mm-thick CsI(Tl) crystals coupled to individual silicon photodiodes that are used to convert the light output of the crystals to electrical pulses. Digital software logic is used to process the signals and create images instead of analog Anger positional circuits. In the three-detector system, the detector heads are positioned at 67.5 degrees between heads, as shown in Figure 9-2. Heads are allowed to be moved in and out (closer to or farther away from the patient). For imaging, the patient sits on a chair with his arms placed on an armrest above the detectors. Data acquisition is typically accomplished in 7.5 minutes by rotating the patient chair by 67.5 degrees, producing a total acquisition arc of 202.5 degrees. With this system, the manufacturer reports a reconstructed spatial resolution of 8.95 mm (at a 20-cm orbit radius) and a sensitivity of 234 cpm/μCi, using the system’s cardiac collimator and a three-dimensional (3D) version of the ordered subset expectation maximization (OSEM) approach for reconstruction. Images of a patient with an inferior-wall defect acquired with this system are shown in Figure 9-3. These systems are now used clinically in several sites, and preliminary reports have been published comparing their performance to that of a standard dual-headed camera when it was found that similar quality could be obtained with a 38% reduction in acquisition time.⁸

Figure 9-1 Upright patient position on the Digirad Cardius 3 XPO triple-head, pixilated detector camera (left) and photograph of the camera (right).

(Images courtesy of Digirad Corp., San Diego, CA.)

Figure 9-2 Geometry of the Digirad Cardius 3 XPO camera. Detectors remain fixed while patient is rotated through 202.5 degrees in a rotating chair.

(Images courtesy of Digirad Corp., San Diego, CA.)

Figure 9-3 Typical SPECT scans obtained with Cardius 3 (C 3) XPO. A technetium-99m-sestamibi study of a patient with an inferior wall defect imaged with the C 3 XPO camera. Sex: male, Height: 5 feet 11 inches. Weight: 240 pounds. Age: 73. Stress: adenosine. Acquisition time: rest, 8.6 minutes; stress, 6.9 minutes. Dose: rest, 10.2 mCi (sestamibi); stress, 32.3 mCi (sestamibi).

(Images courtesy of Digirad Corp., San Diego, CA.)

Preliminary data have recently shown that the acquisition time can be further reduced with this scanner by the application of optimized image reconstruction protocols developed by Digirad. The nSPEED reconstruction⁹ models the depth-dependent detector spatial response of the SPECT systems with a 3D version of the OSEM reconstruction method. In preliminary reports,¹⁰ the image quality improvement with nSPEED, compared to a conventional two-dimensional OSEM technique,¹¹ enables the reduction of acquisition time by 50% while maintaining image quality and information.

CardiArc

CardiArc Limited (Canton, MI) has developed a dedicated nuclear cardiology SPECT camera in which the detector and collimator are redesigned and optimized specifically for cardiac imaging.¹² This device has no visibly moving parts but has a single internally moving part, which is hidden from the patient.¹³ Therefore, from the outside, the detector appears motionless, and for comfort the patient can be positioned upright. Scan times reported by the company are as short as 2 minutes.¹² The camera system and a typical patient position are shown in Figure 9-4. This system was originally designed to use arrays of CZT crystals as detectors. However, owing to the high cost of CZT material and potential long-term stability issues with CZT,¹⁴ the detector design was changed to enable commercial production. Figure 9-5 illustrates the design and the principle of operation of the current model. The system incorporates a high-resolution detector with three curved sodium iodide–thallium (NaI[Tl]) crystals with graduated grooving technology and an array of 60 photomultiplier tubes arranged in three rows (see Fig. 9-5A). The gantry uses a proprietary digital process developed by CardiArc that replaces the conventional Anger logic. Horizontal photon collimation in each slice is accomplished using a thin, curved lead sheet with six narrow vertical slots (aperture arc; see Fig. 9-5A). Vertical slice collimation is accomplished using a series of stationary lead vanes that are stacked vertically between the aperture arc and the NaI(Tl) crystals (see Fig. 9-5B). In this way, data are collected as multiple 1 mm–thick slices using the six vertical apertures to collimate photons so they are detected continuously across the detector surface, with no overlap of data from different apertures. During acquisition, the aperture arc rotates to acquire data from multiple projections, providing 1280 angular samples in 0.14-degree increments over 180 degrees, which is an order of magnitude greater than the conventional camera angular sampling (typically 3 degrees). All detector pixels are active simultaneously, and photons can be detected from multiple angles (see Fig. 9-5C). The movement of the aperture arc is synchronized electronically with the areas of the NaI(Tl) crystals that are imaging the photons passing through the individual slots. The aperture arc’s weight of 35 pounds is lighter than traditional moving gantries, which facilities motion control. The aperture arc movement ranges 9 inches to cover the entire cardiac field of view, and each traverse of the arc takes 10 seconds.

Figure 9-4 Photograph of the CardiArc SPECT-HD demonstrating patient positioning for optimal cardiac imaging and the technologist in the operating position, taking advantage of the built-in radiation shielding.

(Figure from CardiArc Ltd., with permission.)

Figure 9-5 Design and principle of operation of CardiArc camera. The camera uses three stationary sodium iodide–thallium crystals and corresponding photomultiplier tubes for photon detection (A). The aperture arc has six slots (apertures) for horizontal collimation and continuously rotates while imaging. Vertical collimation is achieved by a stack of stationary thin lead vanes, individually separated by foam, that are set between the aperture arc and the detector crystals (B). All detector pixels are utilized simultaneously, allowing imaging of multiple angles (C).

(Figure from CardiArc Ltd., with permission.)

SPECT reconstructed spatial resolution values (full-width half-maximum) quoted by CardiArc range from 3.6 mm (at 82 mm source-to-aperture arc distance) to 7.8 mm (at 337 mm source-to-aperture arc distance). An independent evaluation concluded that the CardiArc system appears to gain image quality by a factor of 5 to 10 when compared to the conventional dual-head camera.¹⁵ A comparison of patient imaging capabilities is shown in Figure 9-6.

Figure 9-6 Stress images of a 55-year-old male, 6 feet 5 inches tall, 235 pounds, with substernal chest pain and shortness of breath. Father had CABG at the same age. Cholesterol = 240. Developed 1.5 to 2.0 mm ST depression in inferior leads at stress, injected with 31.2 mCi technetium-99m-sestamibi at peak stress. Image set (A) was acquired with a conventional dual-head scintillation camera using low-energy high-resolution collimators for 10.6 minutes, and demonstrated an inferobasal defect. Image set (B) was acquired with the CardiArc SPECT-HD for 4.7 minutes and correctly demonstrated a more severe and more extensive defect in the inferior wall to the apex. Angiography revealed high-grade, proximal PDA stenosis.

(Figure from CardiArc Ltd., with permission.)

Spectrum Dynamics

Spectrum Dynamics (Haifa, Israel) has manufactured a system called D-SPECT. The design and principle of its operation are shown in Figure 9-7. The patient is imaged in either a semireclining position with the left arm placed on top of the camera (see Fig. 9-7A) or in the supine position. Acquisition time as short as 2 minutes has been reported.¹⁶ This system uses pixilated CZT detector arrays (see Fig. 9-7B) mounted in nine vertical columns and placed in a 90-degree gantry geometry (see Fig. 9-7C). While CZT detectors are higher in cost, they have advantages of superior energy resolution (by a factor of approximately 1.7 at 140 keV) and compact size compared to the combination of NaI(Tl) with photomultiplier tubes of the conventional Anger camera. With D-SPECT, each detector column is fixed in a mechanical mounting, and the data acquisition is performed by rotating these multiple columns in synchrony. The photons from a given location are detected at multiple angles by multiple columns as the fields of view of the detectors are swept through the region of interest. Each column (see Fig. 9-7B) consists of an array of 1024 CZT elements (2.46 × 2.46 × 5 mm thick) arranged in a 16 × 64 element array with an approximate size of 40 × 160 mm. Each column is fitted with square parallel-hole high-sensitivity collimators, such that the dimensions of each hole are matched to the size of a single detector element. The collimators are fabricated from tungsten to eliminate the production of lead x-rays that might interfere with technetium-201 (²⁰¹Tl) imaging. The collimators have a larger effective diameter than conventional low-energy high-resolution collimators used with scintillation cameras, yielding a significant gain in their geometric efficiency. The collimator has a hole length of 24.5 mm, with a 2.46-mm pitch and 2.26-mm hole diameter. The compensation for the loss in geometric spatial resolution that results from this design is accomplished by the software compensation methods. All data are collected in list mode. A proprietary Broadview iterative reconstruction algorithm based on the maximum-likelihood expectation maximization (MLEM) approach, with resolution recovery and use of the cardiac shape priors, has been developed and patented by the manufacturer.¹⁷ It allows the recovery of image resolution to 5 mm in line source experiments.

Figure 9-7 D-SPECT camera. Photograph of the D-SPECT camera showing patient position (A). A diagram of a single detector column from the D-SPECT camera (B), and a photograph of 9-detector columns configuration (C).

(Photograph courtesy of Spectrum Dynamics, Haifa, Israel.)

Data acquisition is accomplished in a two-step process. First, a 1-minute pre-scan is performed to identify the location of the region of interest. Scan limits and timings are then set for each detector column, and the final scan is performed with each detector column rotating within the limits set from the pre-scan data. This process is shown diagrammatically in Figure 9-8. This process is termed region-of-interest-centric scanning by the manufacturer, because the scan field is limited to only the myocardial region. It has not been possible to measure an absolute value of sensitivity for this system as prescribed by the National Electrical Manufacturers Association quality-control standards,¹⁸ because the sensitivity is significantly dependent on the field of view, defined individually for each patient by the pre-scan process. However, the most centrally located point has been reported to demonstrate a sensitivity of 1407 counts/μCi/min compared to the 160 to 240 counts/μCi/min range generally observed with standard cameras.¹⁹ A case example showing image quality on both D-SPECT and A-SPECT, obtained with the same isotope injection, is shown in Figure 9-9. In a recently published study, when D-SPECT was compared to A-SPECT, the myocardial count rate (with the same injection of the isotope) was 7 to 8 times higher for D-SPECT (Fig. 9-10).¹⁶ Preliminary work has shown that simultaneous dual-isotope SPECT MPI with this camera is feasible using ²⁰¹Tl and technetium-99m, taking advantage of the improved energy resolution of CZT.²⁰

Figure 9-8 The ROI-centric technique utilized by the D-SPECT camera to optimize data collection from the myocardium.

Figure 9-9 A study with standard dual-head SPECT camera (A-SPECT) and D-SPECT. Gender: Male, Age: 61 years, Weight: 200 lbs. Patient had history of coronary disease LAD stent, atypical angina, shortness of breath, diabetes, hypertension, or current smoking. The rest/stress MIBI protocol was performed with rest 8.2 mCi, and stress 37 mCi dose. X-ray angiography found proximal to mid-LAD 70% long lesion correlation. Both A-SPECT and D-SPECT correlate to coronary angiography, but D-SPECT shows more ischemia correlating better to coronary angiography. The acquisition times for both stress and rest were: ASPECT: rest 17 minutes, and stress 15 minutes; D-SPECT: rest 4 minutes, stress 2 minutes.

(Images courtesy of Dalia Dickman, Spectrum Dynamics, Haifa, Israel.)

Figure 9-10 The higher system sensitivity of the D-SPECT system is demonstrated by a significantly higher myocardial count rate (7 to 8 times), compared with conventional SPECT at stress and rest images. CPM, Counts/min.

(From Sharir T, Ben-Haim S, Merzon K, et al: High-speed myocardial perfusion imaging: Initial clinical comparison with conventional dual-detector Anger camera imaging. JACC Cardiovasc Imaging 1:156–163, 2008. Reproduced with permission.)

The higher sensitivity of this system has been exploited to develop new clinical protocols. Cedars-Sinai Medical Center has reported the routine clinical use of this scanner in more than 400 patients using a stress ²⁰¹Tl (2 mCi)/rest tetrofosmin or sestamibi (8 to 10 mCi) protocol. Using half of the radioactivity associated with standard dual-isotope procedures, this protocol includes upright and supine immediate poststress images of 6 minutes each followed by rest injection and immediate 4-minute rest imaging. The total imaging time of this protocol is 19 minutes. Good to excellent image quality without significant extracardiac interference was observed in over 96% of the cases.²¹

Multipinhole Systems

An alternative approach to standard parallel-hole collimation is multipinhole tomography. Previously, promising results with the multipinhole SPECT systems have been presented by several groups in small-animal imaging, surpassing the spatial resolution and detection efficiency of parallel-hole collimation.^22–24 Because all views are active throughout the entire image acquisition period, SPECT imaging in this technique is accomplished without the need for motion of the detector, collimator, or patient. Since image acquisition is accomplished without any electromechanical motion, the manufacturing and servicing costs could potentially be significantly reduced.

The design of the cardiac system for human imaging was reported by Funk et al., with two 9-pinhole collimators attached to the standard dual-headed gamma camera,²⁵ forming, in effect, an 18-pinhole imaging system. This cardiac imaging technology utilizes pinhole collimation to project a simultaneous set of images onto one or more large-area detectors. The principle of operation and a photograph of the actual prototype systems are shown in Figure 9-11. In experiments with the first prototype system, the authors found that spatial resolution of the 9-pinhole collimator with 8-mm diameter pinholes was 30% poorer than that for the parallel-hole collimator. However, the detection efficiency was increased by more than ten-fold. These data allowed them to predict that in comparison to a standard gamma camera, a five-fold increase in sensitivity could be achieved with this technology without degradation of image resolution. Similar increase in sensitivity has been demonstrated by the same group in the small-animal full-ring multipinhole SPECT system.²⁶

Figure 9-11 Principle of operation (A) and photograph of the prototype multipinhole collimator (B) mounted on standard SPECT camera.

(Images courtesy of Dennis Kirch, Nuclear Cardiology Research, Denver, CO.)

With the use of a stationary detector, patient motion–induced inconsistencies between views are eliminated because all views are acquired simultaneously, and multiple isotopes can be imaged simultaneously. All views are active over the entire image acquisition period, leading to a consistent data set for the reconstruction by iterative reconstruction techniques, similar to the acquisition obtained in the full-ring positron emission tomography (PET) system. Correction for patient respiratory motion could also be potentially more reliable compared to standard systems, since all the data acquisition is performed simultaneously from multiple angles.

Multipinhole design potentially suffers from some limitations that will need to be addressed. The approach may be prone to greater formation of artifacts, because it inherently produces an incomplete tomographic data set, and it acquires images from only limited views.²⁵ Background activity from other organs may not be seen in all of the views, which could lead to inconsistencies in the reconstructed data. It is also known that the resolution and sensitivity of pinhole collimators decrease with the distance from the collimator²⁷; however, resolution recovery can be applied during the reconstruction to compensate for this spatial variation. Nuclear Cardiology Research (NCR), located in Denver, Colorado, is currently commercializing this design. Original acquisition images are shown in Figure 9-12, and the reconstructed views are shown in Figure 9-13. The result of breath-by-breath motion correction is shown in Figure 9-14.

Figure 9-12 Original ungated raw views of myocardial perfusion obtained on a 9-pinhole system.

(Images courtesy of Dennis Kirch, Nuclear Cardiology Research, Denver, CO.)

Figure 9-13 Tomographic slices are reconstructed from pinhole views using iterative reconstruction techniques (A) resulting in comparable image quality to the standard MPI reconstruction (B).

(Images courtesy of Dennis Kirch, Nuclear Cardiology Research, Denver, CO.)

Figure 9-14 Multipinhole SPECT images shown before and after motion correction prior to (A) and after (B) reconstruction. Note that the myocardial outline appears slightly smaller, and the interior of the cardiac chamber is better defined.

(Images courtesy of Dennis Kirch, Nuclear Cardiology Research, Denver, CO.)

Astonish

Philips (Milpitas, CA) has developed a fast SPECT reconstruction algorithm (Astonish) that includes corrections for the major factors degrading SPECT image quality. It is based on the OSEM reconstruction method, with built-in noise reduction methods during the iterative process and incorporating corrections for photon scatter, photon attenuation, and variations in spatial resolution. Correction for Compton scattering in the patient improves lesion contrast and is required for accurate attenuation correction. Correction for photon attenuation provides a more accurate representation of the counts from lesions that are at different depths inside the patient. Correction for variations in spatial resolution with depth allows the preservation of sharper details and small lesions with greater conspicuity. The company has developed this approach to shorten the MPI acquisition time without compromising the image quality.

The corrections for variations in spatial resolution use measurements of the changes in spatial resolution with distance from the collimator, as is shown in a general case in Figure 9-15. Calibrations for each of the collimators are measured initially by the manufacturer. Astonish software incorporates this collimator information into both the backprojection and the forward projection parts of the reconstruction. The resolution recover correction in Astonish can be performed with or without attenuation correction.

In Astonish, the corrections for the photon scatter are performed by the ESSE method described by Kadrmas et al.³² The corrections for the photon scatter are performed prior to the attenuation correction in each iterative OSEM step. Corrections for attenuation are performed during the forward projection process. Attenuation correction requires knowledge of both the photon attenuation coefficient and the density of each pixel that the “counts” are forward projected through. The density information is accessed in an attenuation map, modified from a previously acquired density image, with either a scanning line source or from a CT scanner.

One problem that all iterative methods must address is amplification of statistical noise during the reconstruction process. Astonish uses a proprietary (patent pending) noise-reduction method of smoothing both the estimated projection data and the measured projection data internally during the reconstruction process.^33,34 This modification of OSEM allows for optimized control of Poisson image noise while maintaining higher image resolution. In addition, a Hanning prefilter (or optionally no filter) is used to smooth the projection images prior to starting reconstruction. The estimated projections are also smoothed with the same filter prior to the measured/estimated comparison being taken during each subset. This approach can be compared with other methods that smooth the image data after the reconstruction process.³⁵

The clinical performance of Astonish has been tested in a multicenter trial consisting of 221 patients, and preliminary results have been reported.³⁶ Image quality, interpretative certainty, and diagnostic accuracy were evaluated by three “blinded” readers for standard reconstruction for full-time FBP, full-time Astonish (FTA) (64 projections, 20 to 25 sec/projection), and half-time Astonish (HTA) (32 projections, 20 to 25 sec/projection). The HTA data were obtained from the full-time data by using half of the original projections. Stress and rest perfusion image qualities (excellent/good) were 87.8%/83.3% (FBP), 97.7%/95.9% (FTA), and 96.8%/95.5% (HTA), respectively (P < 0.001). Interpretive certainty and diagnostic accuracy (Fig. 9-17) were the same for all these techniques. An example of MPI image quality achievable with Astonish is shown in Figure 9-18.

Figure 9-17 The sensitivity, specificity, and normalcy were determined for filtered backprojection (FBP), full-time Astonish (FTA), and half-time Astonish (HTA).

(From Van Laere K, Koole M, Lemahieu I, Dierckx R: Image filtering in single-photon emission computed tomography: Principles and applications. Comput Med Imaging Graph 25:127–133, 2001. Reproduced with permission.)

Figure 9-18 Example of image quality with Astonish. Stress (A) and rest (B) images are reconstructed with standard filtered backprojection (FBP) image reconstruction, half-time Astonish reconstruction (HTA), and half-time Astonish reconstruction with attenuation correction (HTA-AC). Images of a 209-pound male were acquired with a technetium/technetium protocol with 64 projections and 25 seconds per view full-time and with 32 projections and 25 seconds per view half-time, acquired on CardioMD Philips camera.

(Images courtesy of Philips Healthcare.)

GE Healthcare’s Evolution Software

GE Healthcare (Waukesha, WI) has also developed a modification of the OSEM algorithm that incorporates resolution recovery named “Evolution for Cardiac” or OSEM-RR. Their approach includes modeling of the integrated collimator and detector response function (CDR) in an iterative reconstruction algorithm, and performs image resolution recovery³⁷ based on these parameters. This technique has been described in detail in a recent publication by DePuey et al.³⁸ The OSEM-RR modeling includes basic collimator geometric response function for round-hole-shaped collimators^31,39 that can be applied with good approximation to hexagonal holes.

The CDR compensation technique utilized in OSEM-RR was developed at the University of North Carolina, Chapel Hill, and Johns Hopkins University by Tsui et al.^37,39,40 It is accomplished by convolving the projected photon ray with the corresponding LSF during iterative projection and backprojection. The following parameters are accounted for and compensated: collimator hole and septa dimensions, intrinsic detector resolution, crystal thickness, collimator-to-detector gap, and projection-angle-specific center of rotation–to-collimator face distances. These collimator-specific data are embedded in the software in the form of lookup tables. Some of the relevant acquisition parameters (such as object-to-collimator distance) are obtained directly from the raw projection data.

Additionally, similar to other optimized reconstruction methods, OSEM-RR incorporates noise suppression, which is required since the resolution recovery during the iterative reconstruction process amplifies noise, which can lead to the formation of hot spots in the final image. A maximum a posteriori technique⁴¹ is incorporated to control image noise in the OSEM-RR design. A modified one-step-late algorithm with a Green prior⁴² is utilized. The specific parameters in these reconstructions are optimized separately for each clinical protocol and separately for gated and attenuation-corrected images. The last iteration is performed using a median root prior.⁴³

Siemens Medical Solutions

Siemens has recently released its software (Flash3D) incorporating iterative fast OSEM reconstruction with 3D resolution recovery, 3D collimator and detector response correction, and attenuation and scatter compensation.⁴⁴ SPECT cardiac acquisition protocols (CardioFlash) have been developed utilizing Flash3D, where the acquisition time can be reduced to between 33% and 50%, compared to the standard acquisition protocols with FBP reconstruction. An example of the image quality obtained with CardioFlash is shown in phantoms (Fig. 9-19) and in clinical images (Fig. 9-20). To date, it has been shown in phantom data in combination with a few clinical scans that Flash3D allows faster acquisition protocols but still provides sufficient myocardial uniformity and lesion detectability.⁴⁴ Prospective clinical studies are currently being conducted to assess the performance of this technique with a protocol using 6-degree angular sampling, thus reducing acquisition time by 50%.

Figure 9-19 Example images comparing a standard protocol and a protocol using Flash3D in 36% of the acquisition time. The inferolateral artifact due to lack of attenuation correction is well visible in phantoms.

(From Alenius S, Ruotsalainen U: Bayesian image reconstruction for emission tomography based on median root prior. Eur J Nucl Med Mol Imaging 24:258–265, 1997. Reproduced with permission.)

Figure 9-20 Example patient data acquired with CardioFlash. For images (A) and (C), the original projection data are used but reconstructed with filtered backprojection and Flash3D. The projection data in (B) are extracted from the original data and represent a “what-if” protocol data set with twice the angular step size and 80% dwell time reduction, and reconstructed with Flash3D.

(From Alenius S, Ruotsalainen U: Bayesian image reconstruction for emission tomography based on median root prior. Eur J Nucl Med Mol Imaging 24:258–265, 1997. Reproduced with permission.)

The computer reconstruction times of the 2007 release of Flash3D (Siemens OSEM reconstruction with 3D distance-dependent resolution recovery and optional scatter and attenuation corrections) were improved, and it is now possible to process an entire clinical gated cardiac data set in less then 1 minute on a standard workstation. High correlation (r² > 0.97) has been shown between the ejection fractions obtained from conventional FBP-based protocol and the CardioFlash reconstructions in a preliminary study.⁴⁵

UltraSPECT Wide Beam Reconstruction

UltraSPECT Limited (Haifa, Israel) has developed a standalone workstation (Xpress.cardiac) that utilizes the patented wide beam reconstruction (WBR) algorithm.⁴⁶ The WBR reconstruction technique, phantom validation, and its clinical application have been recently described by Borges-Neto et al.⁴⁷ This solution is available as an additional workstation and can reconstruct data from most existing gamma cameras with standard collimator design. Similar to other methods, WBR models the physics and geometry of the emission and detection processes and attempts resolution recovery. During the iterative reconstruction, it corrects reconstructed voxels, given the information regarding the collimator’s geometry (such as the dimensions and shape of holes or the septa thickness). Furthermore, the detector’s distance from the patient is considered. New gamma cameras provide this information automatically for each angular position from their body contour tracking systems used for orbit prescription. However, this distance can also be obtained by image-processing techniques and definition of the 3D patient body contour, from which the distance of the detector to the body at a given location can be calculated.⁴⁷ In addition, this technique applies statistical modeling of the expected photon emission in order to suppress Poisson noise. The standard approach to overcome noise is the application of Fourier domain post filtering; however, Fourier filtering does not distinguish between the actual signal and noise at a given frequency. WBR instead regularizes the likelihood objective function with a combination of the Poisson and Gaussian distributions. This is accomplished by Fourier analysis of a projection to determine the approximate signal-to-noise ratio that is present in the acquired data and the selection of an optimal noise model to yield the appropriate balance between resolution and noise. An example of WBR image quality is shown in Figure 9-21.

Figure 9-21 SPECT scans of a 56-year-old male hypertensive smoker with no prior history of coronary disease. Images obtained with technetium-99m-sestamibi (dose: 8 mCi at rest and 32 mCi at stress) protocol and acquired with dual-head scintillation camera without attenuation correction. Images were reconstructed with full-time ordered subset expectation maximization (OSEM) (15 minutes rest, 12 minutes stress) (A) and with separate wide beam reconstruction (WBR) (9 minutes rest, 7 minutes stress) acquisitions (B) following the rest and stress OSEM acquisitions, respectively. The actual acquisition time for “half-time” WBR is slightly longer than half as a result of the dead time associated with gantry rotation. However, the WBR acquisition time per camera stop is one-half that for OSEM. Both WBR and OSEM images show the same small apical defect, which is likely physiologic apical thinning.

(Images courtesy Dr. Gordon DePuey, Columbia University, New York, NY.)

Motion-Frozen Reconstruction

Another development related to the quality and therefore potential time reduction in image reconstruction is the “motion-frozen” processing of gated cardiac images, which eliminates blurring of perfusion images due to cardiac motion.⁴⁸ This technique applies a nonlinear, thin-plate-spline warping algorithm and shifts counts from the whole cardiac cycle into the end-diastolic (ED) position. The “motion-frozen” images have the appearance of ED frames but are significantly less noisy, since the counts from the entire cardiac cycle are used. The spatial resolution of such images is higher than that of summed gated images. This technique has been successfully applied to SPECT and PET images. Figure 9-22 shows an example of a SPECT image reconstructed with motion-frozen technique. A significant improvement in image resolution can be observed when compared to standard summed images. The combination of such advanced approaches dedicated to cardiac imaging and the general advances in image reconstruction described could result in further gains in image quality.

Figure 9-22 Short-axis and vertical long-axis of motion-frozen (MF) reconstruction and standard summed reconstruction (SUM) of gated SPECT images. Motion-frozen perfusion images compared to the summed perfusion images in the case of double-vessel disease confirmed by angiography (100% LAD occlusion and 80% LCx occlusion). Both standard quantification technique and visual analysis of summed data identified only the LAD lesion; the additional LCX lesion was identified only by the motion-frozen quantification.

(From Slomka PJ, Nishina H, Berman DS, et al: “Motion-frozen” display and quantification of myocardial perfusion. J Nucl Med 45:1128–1134, 2004. Reproduced with permission.)