Evaluation of Novel Whole-Body High-Resolution Rodent SPECT (Linoview) Based on Direct Acquisition of Linogram Projections

MATERIALS AND METHODS

Linoview SPECT Device

The animal Linoview SPECT device (Linoview Systems; Fig. 1) is based on four 2- by 5-in. γ-ray detectors (Rayvision). Ten 1- by 1-in. position-sensitive photomultiplier tubes per detector (R8520-00-C12; Hamamatsu Photonics) are coupled to a pixelated CsI(Na) scintillator (5-mm thickness, 21 × 52 pixels of 2.5 × 2.5 mm). The detectors are connected to a computer via an analog-to-digital board (PCI-416; Datel) driven by the acquisition software KmaxNT (Sparrow Corp.). The intrinsic spatial resolution is 2.5 mm, the intrinsic energy resolution at 140 keV is 35%, and the intrinsic sensitivity in an energy window of 35% width centered on the photopeak is 42%. The collimator and detector housings are made of tungsten alloy (W-Ni-Fe; W, 95%; density = 18; thickness, 7 mm [XiAn Tungsten & Alloy]). The rake collimator has a slit aperture parallel to the longitudinal direction (Fig. 2A). This tunable slit aperture is composed of 2 square iridium rods (2 × 2 × 60 mm) inserted in 2 linear sliding holders made of tungsten alloy (Fig. 2B), allowing a slit width of 0–5 mm. The distance between the slit aperture and the scintillator surface is 62 mm. Longitudinal collimation is achieved by 20 parallel trapezoidal tungsten sheets (55 mm high, 125-mm bottom width, 16-mm top width, 0.5-mm thickness, 2-mm interspace) sandwiched in a nonattenuating medium (extruded polystyrene) (Fig. 2C). The radial and orbital motions are managed by linear slide guides (MNN18-G3; Sheenberger) and stepper motors acting on an endless screw (L1MGB; EAD Motors), driven by a controller board (DCB-261; AMS). Mechanical limit switches serve as reference points. All linear guides and stepper motors are mounted on an aluminum plate with a position accuracy of 0.1 mm.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1.

Linoview SPECT device (Linoview Systems), consisting of 4 pixelated CsI(Na) detectors (A) associated with rake collimator (B). Orbital and radial motions are obtained with stepper motors (C and D) acting on endless screws (E and F).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2.

Collimators. (A) Upper view of collimator with adjustable slit aperture parallel to longitudinal direction. (B) Lateral view of 2 square iridium rods inserted in tungsten holders. (C) Inferior view of collimator showing tungsten sheets (D) sandwiched in extruded polystyrene (E).

The physical dimensions needed for the tomographic reconstruction are the focal length of the collimators, the distance from the slit aperture to the axis origin, the transverse position of the slit aperture on the crystal, and the slit aperture position along the orbit. All of these parameters were obtained by a calibration procedure consisting of 3 tomographic acquisitions of a linear source located approximately at the transverse coordinates (−5, −5 mm), (+5, −5 mm), and (−5, +5 mm). A numeric program automatically computes the actual source positions and physical dimensions of the system (Appendix). The benefit of this procedure is to avoid difficult direct measurements of distances. Furthermore, it is automatic and provides the effective dimensions of the system, taking into account γ-ray penetration into the edges of the iridium rods and into the crystal. As a result, the computed effective focal length is somewhat greater than that obtained by direct measurements.

RESULTS

Whole-Body Phantom

With a 0.20-mm collimator aperture, all of the 0.75-mm hot rods were clearly distinguishable on the whole slice (2 mm thick), whereas with a 0.1-mm aperture, the hot rods separated by 0.35 mm were distinguishable up to a 10-mm depth (Fig. 3). The active ring surrounding the 0.75-mm hot-rod phantom was the activity located between the acrylic disk and the inner wall of the phantom housing.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 3.

Transverse slices of two 30.5-mm-diameter hot-rod phantoms filled with ^99mTc; acquisition statistics were 32 MBqh (left) and 57 MBqh (right). Distances between rod edges were 0.75 mm (left) and 0.35 mm (right). This distance determines spatial resolution achieved.

Brain Phantom

The 0.50-mm hot rods were clearly distinguishable on the whole slice (2 mm thick), whereas the hot rods separated by 0.35 mm were clearly distinguishable up to a 10-mm depth in the phantom and remained distinguishable in the central region, although the image was somewhat blurred (Fig. 4). Some activity was present between the phantom housing and the acrylic disks that were obtained by manual cutting.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 4.

Transverse slices of two 22.5-mm-diameter hot-rod phantoms filled with ^99mTc; acquisition statistics were 34 MBqh (left) and 111 MBqh (right). Distances between rod edges were 0.50 mm (left) and 0.35 mm (right). This distance determines spatial resolution achieved. Top slices represent enlargement (factor 3) of central region of phantoms.

31.5-mm-Diameter Cold-Rod Phantom

With a 0.2-mm collimator aperture, both the 2.5-mm and the 1.3-mm cold rods were clearly visible on the whole slice (10 mm thick) of the phantom, whereas the 0.65-mm rods were clearly visible up to a 12-mm depth in the phantom and remained visible in the central region, with some blurring (Fig. 5).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 5.

Transverse slices of three 31.5-mm-diameter cold-rod phantoms filled with ^99mTc; acquisition statistics were 18, 24, and 59 MBqh (from left to right). Diameters of cold rods were 2.50, 1.30, and 0.65 mm (from left to right). Slice at right represents enlargement (factor 2) of central region of 0.65-mm-diameter cold-rod phantom.

^99mTc-DMSA and ¹¹¹In-Pentetreotide Imaging in Mice

Figure 6 shows coronal and transverse slices through mouse kidneys obtained 5 h after injection of ^99mTc-DMSA and 24 h after injection of ¹¹¹In-pentetreotide. Both tracers were concentrated over the cortex area, whereas no activity was seen in the pelvis or pelviureteral junction (∼1-mm diameter). The kidney uptake values calculated from in vivo SPECT (^99mTc, 15.6 percentage injected dose [%ID]; ¹¹¹In, 9.1 %ID) were similar to those obtained ex vivo by γ-counting (^99mTc, 15.2 %ID; ¹¹¹In, 9.6 %ID).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 6.

Coronal (A and C) and transverse (B and D) slices through kidneys of 20-g male mouse imaged 5 h after injection of 22 MBq of ^99mTc-DMSA (A and B) and 22-g female mouse imaged 24 h after injection of 11 MBq of ¹¹¹In-pentetreotide (C and D). For both acquisitions, distance between mouse center and collimator aperture was 16 mm. Note that tracer was concentrated over cortex area, whereas no activity was seen in pelvis or pelviureteral junction (≈1-mm diameter). (E) Similar pattern of activity distribution was seen on ex vivo autoradiogram obtained for mouse injected with ¹¹¹In-pentetreotide.

DISCUSSION

The results obtained with this animal SPECT system indicate a significant improvement in spatial resolution over those obtained with other dedicated small-animal SPECT systems. The improvement is spectacular, especially with cold-rod phantoms. Indeed, high-contrast images were obtained with 1.3-mm cold rods and low acquisition statistics (24 MBqh), whereas in the literature, images obtained with cold rods of the same size usually have had much less contrast, even with high acquisition statistics, such as 333, 1,850, and 2,146 MBqh (2–3,5). With 0.65-mm cold rods and 59 MBqh, all of the rods still were visualized, but with lower contrast, which was particularly marked in the center of the phantom. Different factors, such as the collimator design and direct linogram acquisition, contribute to this improvement.

In rod phantoms in which the radioactivity is spread over an extended area, the spatial resolution depends significantly on the long-range tail of the collimator point spread function (PSF). Iridium was used as the material for the slit aperture instead of a tungsten alloy. Besides uranium, iridium is the most effective material for γ-ray attenuation. Its attenuation coefficient is 39 cm⁻¹ at 150 keV; that for tungsten is 30 cm⁻¹ (20). The increased stopping power results in a faster decrease in the collimator PSF tail. Also, the use of a tunable slit allows very small apertures; when the 2 iridium parts are set in contact with each other at 140 keV, the effective aperture is 0.22 mm as a result of the γ-ray penetration. Furthermore, it was already shown that for a similar collimator spatial resolution, the translation of this spatial resolution in the reconstructed transverse plane with LOrA leads to a spatial resolution better than that obtained with a classical revolution orbit in the center of the reconstructed slice (13).

In the cold-rod phantoms, the active region represents almost the whole phantom volume. The challenge is to distinguish the small cold regions from the reconstruction artifacts generated by the surrounding activity. Thus, the image quality depends strongly on the signal-to-noise ratio. This ratio can be improved by 2 ways: by reducing the noise present in the acquired data and by increasing the noise stability of the tomographic system.

With the classical revolution orbit, or stationary multipinhole system, the positioning of the animal close to the collimator aperture introduces truncation problems in the reconstruction process. The latter problems generally are solved by increasing the radius (≥25 mm); unfortunately, this process leads to a reduction in spatial resolution and sensitivity. In contrast, LOrA allows setting the collimator closer to the animal without missing any data (13). As a result, the optimal efficiency and spatial resolution of the collimator are used without introducing truncation problems and without altering the stability of the reconstruction process. For example, in a mouse brain study, the brain center can be positioned ∼10 mm away from the collimator aperture. In such a setting, a LOrA SPECT system equipped with a 0.89-mm-diameter pinhole collimator would provide a spatial resolution similar to that obtained with a classical SPECT system equipped with a 0.3-mm-diameter pinhole collimator (Fig. 7A). These latter 2 factors—reduced distance from collimator to target and larger collimator aperture—would lead to a theoretic improvement in sensitivity by a factor of ∼60 in favor of a LOrA SPECT system compared with a classical SPECT system (Fig. 7B). In this configuration, a LOrA SPECT system with 4 detectors equipped with pinhole collimators would be equivalent to a 240-multipinhole system.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 7.

Theoretic spatial resolution (A) and sensitivity (B) for pinhole collimator (60-mm collimator focal length, 2.5-mm detector intrinsic spatial resolution) as function of target-to-aperture distance. When imaging small target organ, for example, mouse brain, organ would be positioned much closer to collimator aperture in LOrA SPECT (5–15 mm) than in classical SPECT (20–30 mm). As a result, LOrA SPECT equipped with 0.89-mm-effective-diameter pinhole (dashed line) provides spatial resolution similar to that obtained with classical SPECT and 0.30-mm-effective-diameter pinhole (solid line). These 2 factors, increased hole diameter and reduced target-to-aperture distance, lead to improvement in sensitivity by factor of ∼60 in center of organ. FWHM = full width at half maximum.

Another major feature of the implementation of LOrA is that it directly provides linograms rather than the usual sinograms. Although a point source acquired with a converging collimator along a revolution orbit does not exactly result in a sinusoidal curve but presents some distortion, the following considerations are still valid (21). Considering the limit of infinitely fine sampling, both linogram and sinogram acquisitions provide a complete set of data allowing an exact tomographic reconstruction. In a real study, one must deal with discrete data sampling and reconstruction of a discrete matrix. Tomography is known to be an ill-posed problem resulting in an amplification of the noise present in the acquired data during the reconstruction process (4). However, linograms are known to provide data sampling different from that provided by sinograms. This feature is clearly visible in the Fourier space of frequencies, in which linograms provide samples of the Fourier transform on radial points located on concentric equidistant squares (pseudopolar Fourier transform) rather than on radial points located on equidistant circles, as with sinograms (21). Besides the fact that this difference allows faster computation in direct Fourier reconstruction (14), work in progress suggests that, in contrast to sinograms, linograms offer a discrete Radon transform that is algebraically exact and invertible (22). Furthermore, in sinogram reconstruction, the transverse coordinate on the detector plays the role of a radial coordinate, and oversampling of the sinogram on this coordinate allows reduction of the noise present in the reconstructed slice (23). However, this radial oversampling is hampered by the finite spatial resolution of the detector. In contrast, in linogram reconstruction, because the transverse coordinate on the detector is replaced by the coordinate of the detector on the orbit, there is no restriction of the subsequent radial oversampling (13). Further work should be conducted to evaluate whether these properties could improve the stability of the system against the noise present in the acquired data. Such improvement in stability could be an additional explanation for the good results obtained for cold-rod phantoms.

Preliminary studies with mice confirmed the initial phantom results in that a high spatial resolution could be reached with low activities (11–22 MBq); the SPECT images allowed a clear delineation of the radioactivity concentrated over the cortex area of the kidneys, whereas the pelvis and the pelviureteral junction (1 mm) appeared as cold areas. Furthermore, the quantitative data derived from SPECT were similar to those obtained by ex vivo counting in the isolated organs. In addition, in the mouse injected with ¹¹¹In-pentetreotide, the kidney radioactivity distribution seen with SPECT was in agreement with the ex vivo autoradiogram (Fig. 6E).

The use of a rake collimator, which has a fanbeam geometry, ensures the completeness of the acquired data along the whole body length without the need for a sophisticated orbit (10–12); the whole body can be imaged in 1 linear motion of the detectors. This simple linear motion can be achieved in a very short time (∼5 s) compatible with dynamic SPECT after tracer injection. For this purpose, the tunable slit can be opened up to 5 mm to increase sensitivity, when needed. The scan range in the longitudinal direction was 5 cm in this prototype but could be increased; 7.5 cm should be appropriate for whole-body studies with mice, and 15 cm should be appropriate with rats. The current rake collimator was designed for whole-body pharmacokinetic studies and provided a 2-mm thickness for reconstructed slices. Multibed acquisitions—with a bed shift of 0.5 mm between acquisitions—provided high-quality coronal images of small organs, such as the kidneys. In the future, longitudinal spatial resolution recovery for multibed acquisitions should be implemented in the reconstruction program and evaluated. Further, the use of a slanted rake collimator also focusing in the longitudinal direction or a pinhole collimator should be evaluated for dedicated organ studies.

The major drawback of the current prototype is the flatness of the γ-detector used. As a consequence, the distance from the aperture to the detector edge is √2 longer than the distance to the detector center, reducing the local sensitivity by a factor of ∼2. Also, at the edge of the detector, the γ-rays reach the crystals at an incident angle of 45°, leading to an alteration in the spatial resolution. Indeed, because of γ-ray penetration into the crystals, a signal will be generated not only in the crystal directly hit by the γ-rays but also in adjacent crystals. The resulting spatial resolution deterioration already noted with ^99mTc is much more pronounced with high-energy radionuclides, such as ¹¹¹In. In the future, a cylindric design of the γ-detector would overcome the 2 problems described here. In such a design, the increase in crystal thickness would further improve the system sensitivity without altering the spatial resolution.

Because of the low-energy resolution of the system, the scattered γ-rays cannot be rejected and will alter to some extent the image resolution. Using a system equipped with a rake collimator and a detector with similar pixel size and energy resolution, Brasse et al. showed that ∼20% of the detected photons were scattered inside the mouse and 1% were scattered inside the collimator (16). Therefore, in the future, the system also should benefit from the use of cadmium-zinc-telluride (CZT) detectors, which offer better energy resolution.

CONCLUSION

Our studies showed that this new animal SPECT, based on linogram acquisitions, provides high-spatial-resolution images of cold and hot structures in the submillimeter range. The preliminary mouse studies showed that this system will be most useful for in vivo high-resolution SPECT and quantitative biodistribution studies with rodents, even with medium-energy radioisotopes that are difficult to image, such as ¹¹¹In.

APPENDIX

The linear sources used for the calibration procedure consisted of a glass capillary with a 1-mm inner diameter filled with either ^99mTc or ¹¹¹In. Acquisitions were performed with the collimator aperture ∼20 mm distant from the center of the gantry.

In a linogram acquired for slice z of source s, the coordinate (V_α) of detector α along its linear orbit was defined as follows: Eq. 1A where r_α is the transverse coordinate of the γ-rays on detector α and s = 0,1,2 for the linear source positioned approximately at the coordinates (−5 mm, −5 mm), (5 mm, −5 mm), and (−5 mm, 5 mm) against the center of the gantry and parallel to the collimator aperture. As shown in the system view depicted in Figure 1, the horizontal x-axis is pointing to the right, the vertical y-axis is pointing to the top, and the longitudinal z-axis is parallel to the collimator aperture. The orbit coordinate V_α increases when the detectors move in the clockwise direction. The r axis on detector α points in the direction of the clockwise motion. The detectors are numbered from 0 to 3 in a clockwise fashion starting from the right detector (Fig. 1). The parameters A_αz^s and B_αz^s are obtained simply by performing a linear least-squares fitting of the maximum pixel in each line r of the linograms of Equation 1A.

Considering that the γ-rays cross the collimator aperture, the A_αz^s and B_αz^s parameters are described by the following equations: Eq. 2A Eq. 3A Eq. 4A Eq. 5A Eq. 6A Eq. 7A Eq. 8A and Eq. 9A where the unknowns are U_αz, the distance between the aperture of detector α and the system axis parallel to the detector orbit; F_αz, the effective focal length of detector α; δr_α, the offset of the actual aperture position from the detector coordinate origin; δV_αz, the offset of the actual aperture position from the orbit coordinate V_α; and the source coordinates x_z¹, y_z¹, x_z², and y_z². The values for coordinates x_z⁰ and y_z⁰ can be fixed arbitrarily, and the chosen values will determine the axis system used for the reconstructed space; in our system, the values used were x_z⁰ = −5 mm and y_z⁰ = −5 mm.

From Equation 2A, we obtain Eq. 10A and from Equation 3A, we obtain Eq. 11A

Elimination of δV_αz by subtraction in Equations 6A to 9A provides the following equations: Eq. 12A Eq. 13A Eq. 14A Eq. 15A Eq. 16A Eq. 17A Eq. 18A and Eq. 19A

Removing the source coordinates between each consecutive even–odd couple in Equations 12A to 19A by use of Equations 10A and 11A provides the following δr_α offsets: Eq. 20A and Eq. 21A

Now, the domino-falling process is initiated. Replacing the δr_α offsets in Equations 12A to 19A with their values in Equations 20A and 21A directly provides the source coordinates x_z¹, y_z¹, x_z², and y_z²; the same operation in Equations 6A to 9A directly provides the δV_αz offsets; subtracting Equations 2A to 5A between sources 1 and 0 removes the unknown U_α and provides the effective focal length F_αz; and replacing F_αz, x_z^s, and y_z^s with their values in Equations 2A to 5A provides the actual distance U_αz used for the acquisitions.

The deviations of the glass capillary used for the linear source, of the square iridium rods, and of the linear sliding guides from a straight line are much smaller than the spatial resolution of the system; the same is true regarding the deviation of the crystal array surface from a flat plane. As a consequence, the variations in the actual parameters δr_αz, δV_αz, F_αz, and U_αz along the longitudinal coordinate z are only attributable to spatial misalignments, mainly of the linear source and the collimator aperture (the detectors are heavy—∼20 kg—and some flexing of the holders cannot be avoided). As a result, the actual parameters δr_αz, δV_αz, F_αz, and U_αz are linear functions of the coordinate z. Thus, a linear fit of the computed parameters is applied to remove the fluctuations induced by the noise present in the acquired data. On the basis of our experience, 3 acquisitions of 30 s each with 7 MBq per millimeter of linear source are sufficient to obtain a reliable calibration of the system.