Initial Investigation of Preclinical Integrated SPECT and MR Imaging

Abstract

Single-photon emission computed tomography (SPECT) can provide specific functional information while magnetic resonance imaging (MRI) can provide high-spatial resolution anatomical information as well as complementary functional information. In this study, we utilized a dual modality SPECT/MRI (MRSPECT) system to investigate the integration of SPECT and MRI for improved image accuracy. The MRSPECT system consisted of a cadmium-zinc-telluride (CZT) nuclear radiation detector interfaced with a specialized radiofrequency (RF) coil that was placed within a whole-body 4 T MRI system. The importance of proper corrections for non-uniform detector sensitivity and Lorentz force effects was demonstrated. MRI data were utilized for attenuation correction (AC) of the nuclear projection data and optimized Wiener filtering of the SPECT reconstruction for improved image accuracy. Finally, simultaneous dual-imaging of a nude mouse was performed to demonstrated the utility of co-registration for accurate localization of a radioactive source.

Introduction

Through the use of highly specific radiolabeled molecular probes, nuclear imaging techniques such as single-photon emission computed tomography (SPECT) can provide insight into a wide range of biological processes with demonstrated applications in neurology, cardiology, oncology and, more recently, stem cell research. However, the relatively poor spatial resolution of radionuclide techniques can make unambiguous localization of the probes extremely difficult, especially when the images lack significant anatomical detail for reference. Limited spatial resolution can also hamper quantification of the probe concentration, especially when localized in small volumes. A unique advantage of SPECT over position emission tomography (PET) is that simultaneous multiple isotope imaging is also possible if the detector has high energy resolution. This opens up the possibility of labeling different molecules with different radioisotopes and performing multidimensional molecular imaging to investigate various biological processes simultaneously.

In contrast to SPECT, magnetic resonance imaging (MRI) can provide exceptionally high spatial resolution anatomical information as well as localized chemical and physical information (i.e. metabolite concentrations, water diffusion characteristics). However, recently developed molecular probes using MR contrast agents have much lower sensitivity (about 10,000-fold) on a ‘per-molecule’ basis than radionuclide techniques. In addition, the nonlinear relationship between probe concentration and the MR signal intensity makes absolute quantification difficult.

SPECT and MRI each have their respective advantages and limitations. Integrating these two modalities in a synergistic manner would allow researchers to exploit the strengths of both techniques. For example, MRI data can be utilized to improve the accuracy and spatial resolution of the reconstructed SPECT images. Such improved resolution should reduce partial volume effects, where inaccuracies in quantification of the radiotracer concentrations can occur in smaller structures. This effect becomes significant, for example, in the interpretation of SPECT images of tumors, where a measured decrease in the uptake of a radiotracer following treatment could indicate tumor shrinkage, a change in biological function, or both. Segmentation of the MR images can be used to facilitate attenuation correction of the nuclear projection data, also improving the accuracy of the SPECT reconstruction. Anatomical MR images can provide a reference for the SPECT images, allowing for improved localization. The ability to acquire SPECT and MRI data simultaneously would open up new research opportunities in dynamic imaging using both SPECT radionuclides and MRI contrast agents at the same time with optimum spatial and temporal co-registration. This would provide motivation for the development of appropriate bi-functional imaging probes. Additional advantages of simultaneous SPECT and MRI measurements over separate or sequential acquisitions include the reduction of co-registration errors, decrease in the overall scan time, and the possibility of using the MR images to correct for motion artifacts in the SPECT data. A comprehensive rationale for combining SPECT with MRI was given by Wagenaar et al (1).

While the integration of SPECT and MRI offers numerous advantages and new opportunities, it also presents many technological challenges. The SPECT detectors must function within an operating MRI scanner. Likewise, the SPECT hardware must not significantly perturb the MR images. Due to these challenges, the development of a combined SPECT and MRI system (henceforth called MRSPECT) is in its infancy, and a very limited amount of research has been reported to date. Breton el at and Goetz et al used a strategy similar to PET-CT systems in which a small animal SPECT system was brought in close proximity to a separate MRI system (2,3). While they demonstrated excellent results, they utilized a substantially low magnetic field (0.1 T) and performed sequential SPECT and MR imaging. Meng et al presented the design of an MR-compatible SPECT system for mouse brain imaging based on cadmium-zinc-telluride (CZT) nuclear radiation detectors (4). While they investigated the effects of the SPECT and MRI components on each other, they confined their study to the use of a ⁵⁷Co point source for SPECT imaging and were not able to acquire simultaneous SPECT and MRI experimental data. We recently reported on the design and operation of an MRSPECT system also based on CZT detector technology (5). In that study, the effects of the SPECT and MRI components on each other were characterized through various phantom experiments. The results demonstrated the feasibility of co-registered, simultaneous SPECT and MR imaging. In this study, we utilized this MRSPECT system to investigate the integration of SPECT and MRI for improved image accuracy in both a phantom and a small animal.

Materials and Methods

Design of the MRSPECT System

For SPECT imaging, we utilized a CZT-based nuclear radiation detector (Gamma Medica-Ideas) that consists of a 25.4 × 25.4 × 5 mm CZT crystal coupled to 256 (16 × 16) detector elements. The detector elements are integrated with low-power application-specific integrated circuit (ASIC) readout electronics which are connected to a carrier board that was mounted inside a plastic box. This housing and cables connecting the carrier board to the interface electronics were wrapped with a fine copper mesh for radiofrequency (RF) shielding. Remaining components such as the power supplies, interface electronics, and computer were located in a control room outside an RF-shielded MRI room.

The 5 mm thickness of the CZT crystal makes the detector suitable for use with isotopes which emit low-energy gamma rays (e.g. absorption efficiency ~90% at 140 keV). Wagenaar et al previously reported that this detector yields a high quality pulse height spectrum (PHS) of a ⁵⁷Co source (121 keV) in up to a 7 T magnetic field, with an energy resolution (~5%) considerably better than that of scintillation-based detectors in the absence of a magnetic field (6). The high energy resolution of the CZT detector allows for much narrower windowing around the photopeak without losing detection sensitivity, resulting in an improved signal-to-noise ratio (SNR). Such energy discrimination also makes it possible to perform multi-isotope studies when using combinations of low energy gamma ray emitters, such as: ^99mTc and ¹²³I; ^99mTc and ²⁰¹Tl; and ^99mTc and ¹¹¹In.

A 2.54 × 2.54 × 2.54 cm lead parallel-hole collimator consisting of 1.2 mm² squares holes separated by 0.4 mm septa was positioned on the CZT detector. While pinhole collimators are necessary for micro-SPECT applications, we selected a parallel-hole collimator to simply the setup and maximize sensitivity in this preliminary study. The open end of the collimator was inserted between the rungs of a custom-built RF birdcage coil, which was utilized for MR imaging (7). Additional lead shielding was placed on the front of the detector housing to minimize the number of stray gamma rays striking the detector. The configuration of the nuclear radiation detector and RF coil is illustrated in Figures 1a and 1b. The overall design avoids any attenuation of the detected gamma rays by the RF coil materials and allows for adjustment of the distance from the collimator to the object being imaged. In this study, this distance was set to 10 mm, which represents a reasonable compromise between SPECT and MR image quality. Shorter distances would slightly improve the SPECT resolution, but substantially increase the adverse affects of the lead components on the MR images. To simply the setup in this preliminary study, the position of the nuclear radiation detector and RF coil remained fixed while the object being imaged was manually rotated in order to acquire data at different views for tomographic imaging.

Figure 1.

(a) End view with a phantom and (b) top view with a mouse of the integrated nuclear radiation detector and RF coil within the MRI magnet bore. (c) Schematic of the Lorentz shift. After interaction of an incoming gamma ray with the CZT crystal, the resulting electron is deflected by a distance Δx due to the Lorentz force.

The nuclear radiation detector and RF coil were placed within a 4 T whole-body magnet (Magnex Scientific) with a clear bore of 940 mm. The large bore size accommodates a variety of instruments, such as in this study, thus making it possible to pursue multimodality imaging research. The MRI system uses a 12 kW 4T12K RF amplifier (CPC) and a whole body gradient coil with a 13-channel shim set (Tesla Engineering), along with a QDCM 950/200/4400 gradient amplifier (MTS) capable of generating fields up to 30 mT/m with a slew rate of 115 T/m/s at 940V/440A. The hardware is interfaced to a MR6000 console (MR Solutions) for pulse sequence generation and MR data acquisition.

Nuclear Imaging Corrections

When a semiconductor-based nuclear radiation detector is placed within a magnetic field, electron-hole pairs created from the interaction of absorbed gammy rays are subject to the Lorentz force. As a result, when the CZT-based detector is placed in any orientation other than parallel or anti-parallel to the main static magnetic field of the MRI system (+/− z-direction), electrons traveling towards the anode will experience a shift in their detected position as illustrated in Figure 1c. The magnitude of this shift depends on the magnetic field strength, detector orientation, absorption and diffusion properties of the semiconductor, bias voltage across the semiconductor, and energy of the incident gamma rays. For our particular MRSPECT system configuration, we previously measured a mean ‘Lorenz shift’ of 1.4 mm (5). This effect should be taken into account prior to SPECT reconstruction by shifting the nuclear projection data to their proper locations.

In addition, the raw images of any nuclear imaging system must be normalized by a flood field image to account for inherent spatial variations in the detector sensitivity. This is particularly important for semiconductor-based detectors, where inherent imperfections in the crystal result in a non-uniform sensitivity across the detector elements. To perform this uniformity correction, a flood field image is first acquired by exposing the detector to spatially uniform activity. Subsequent nuclear projection images are then normalized by this flood field image pixel by pixel to obtain the uniformity-corrected images. To investigate the effects of the Lorentz shift on this uniformity correction, we acquired ^99mTc flood field images using a 10% energy window around the 140 keV photopeak both outside (at 0 T) and inside (at 4 T) the MRI magnet bore with the detector oriented as in Figure 1.

Phantom Experiments

To investigate the integration of SPECT and MRI using our MRSPECT system, we first imaged a phantom consisting of a hollow acrylic cylinder with an inner diameter of 19 mm, shell thickness of 1.6 mm, and length of 70 mm. Smaller acrylic rods each 3 mm in diameter were positioned axially within this cylinder. The interior region was filled with 5 mCi activity of ^99mTc and a 10 mM CuSO₄ solution. The entire phantom was positioned inside the RF coil with its axis parallel to the z-direction (Figure 1a).

The presence of lead components (collimator and shielding) was previously found to deteriorate the homogeneity of the main static magnetic field (5). Thus prior to SPECT/MRI data acquisition, the phantom was shimmed using up to 3^rd order corrective shim coils powered by a MXA-13-R shim power supply (Resonance Research). The optimum shim channel values were calculated using a constrained least-squares-fit algorithm (8). After shimming, simultaneous SPECT/MRI data were acquired for 30 equally spaced views around 360 degrees about the z-axis. For each view, data was only acquired after the phantom was positioned and not during rotation from the previous view in order to avoid any motion artifacts.

For MR imaging, an axial slice through the center of the phantom was acquired using a two-dimensional spin-echo (SE) pulse sequence with the following parameters: repetition time (TR) = 500 ms, echo time (TE) = 20 ms, field-of-view (FOV) = 40 mm², matrix = 128 × 128, slice thickness = 5 mm, receiver bandwidth = 33.3 kHz, and number of excitations (NEX) = 2. Nuclear projection data were acquired using a 10% energy window around the 140 keV photopeak. For each view, the nuclear projection data acquisition time coincided with the length of the MRI scan (~2 minutes).

Image Reconstruction

Prior to tomographic reconstruction, the nuclear projection data underwent uniformity correction. For comparison, one copy of the projection data was normalized using the flood field image acquired outside the magnet while a second copy was normalized using the flood field image acquired inside the magnet. After uniformity correction, compensation for the Lorentz shift was performed by shifting the projection image for each view by 1.4 mm in the appropriate direction. For comparison, a copy of the projection data without any shift correction was retained.

Attenuation correction (AC) of the nuclear projection data was also investigated. Determination of an accurate object-specific attenuation map is required to perform AC. Strategies for obtaining this attenuation map include 1) importing and registering the map derived from another modality, 2) acquiring transmission data for estimating the map using an external source, and 3) estimating the map solely from the emission data. Our integrated SPECT and MR imaging allows us to utilize the first method. The contribution of each (active) voxel to a given projection image was attenuated by the product over N of exp(−μ_Nd_N), where μ_N is the linear attenuation coefficient of the N-th phantom component (water or acrylic) and d_N is the distance across the N-th component region from the voxel to the detector. For each view, the co-registered MR image was used to segment the different component regions, define the known regions of activity (water region), and measure the d_N. An AC factor was then calculated and applied to the projection data. For comparison, a copy of the projection data without AC was also retained.

For each of the various nuclear projection image sets, data corresponding to the location of the MRI slice were selected for SPECT reconstruction. For each set, half a million total counts were selected and corrected for decay across the data acquisition period. Filtered back-projection (FBP) with the Shepp-Logan filter (cut-off frequency = 0.315 cycle/mm) was performed on these data, and the resulting reconstructed SPECT images were interpolated to the same FOV and matrix as the corresponding MR image for direct comparison.

After SPECT reconstruction, we next investigated the utilization of the high-spatial resolution MR image for improvement of the spatial resolution and accuracy of the SPECT image. Following the technique of Chou et al (9), the Wiener filter was applied to the SPECT image with the point spread function of the MRSPECT system modeled by Gaussian normal distribution function characterized by the (unknown) full width at half maximum (FWHM) w. The optimum value of w was determined by maximizing the mutual information between the resulting filtered SPECT image and the co-registered MR image. For comparison, a copy of the SPECT image without any filtering was also retained.

In vivo Animal Imaging

After the phantom experiment, a nude mouse was utilized to test small animal imaging with our MRSPECT system. The animal was first anesthetized and fixed within an acrylic cylinder. A small tube with an inner diameter of 2 mm was filled to a height of 4 mm with 1 mCi of ^99mTc and inserted rectally into the animal. The cylinder containing the animal was then placed within the MRSPECT system for simultaneous dual-imaging (Figure 1b). Shimming was first performed using the previously described procedure, then simultaneous SPECT/MRI data were acquired for 30 views about 360 degrees using the same imaging parameters as with the phantom experiment. Inspection of the MR images across the different views verified that the position of the animal within the cylinder remained fixed throughout the experiment. Based on the results of the phantom experiment, the nuclear projection data was normalized using the flood field image acquired inside the magnet and corrected for the Lorentz shift. SPECT reconstruction was then performed using the previously outlined methodology.

Results and Discussion

Phantom Experiment

The axial MR magnitude image of the phantom is shown in Figure 2a. The corresponding SPECT images reconstructed using the various nuclear projection data sets are shown in Figures 2b–2e. The SPECT image resulting from application of the optimized Wiener filter is shown in Figure 2f. Profiles of the SPECT images of Figures 2d–2f across the gray dashed line in Figure 2a are plotted in Figure 3.

Figure 2.

(a) MR image of the phantom. The gray dashed line was used to generate the profiles in Figure 3. (b) SPECT image reconstructed using projection data normalized by the flood field image acquired outside the magnet. (c) SPECT image reconstructed using the projection data normalized by the flood field image acquired inside the magnet, but not corrected for the Lorentz shift. (d) SPECT image reconstructed using the projection data normalized by the flood field image acquired inside the magnet and corrected for the Lorentz shift. (e) SPECT image reconstructed using the projection data properly normalized and corrected for the Lorentz shift, as well as corrected for attenuation using an MRI-based map. (f) SPECT image reconstructed using the fully corrected projection data and restored using a Wiener filter optimized though the maximization of mutual information with the co-registered MR image.

Figure 3.

Profiles of the SPECT images taken across the gray dashed line in Figure 2a. “Standard” corresponds to the SPECT image reconstructed using the projection data normalized by the flood field image acquired inside the magnet and corrected for the Lorentz shift. “AC” corresponds to the SPECT image reconstructed using the projection data properly normalized and corrected for the Lorentz shift, as well as corrected for attenuation. “Filtered” corresponds to the SPECT image reconstructed using the fully corrected projection data and restored using the optimized Wiener filter.

The SPECT image reconstructed using the projection data normalized by the flood field image acquired outside the magnet is highly distorted (Figure 2b). The SPECT image reconstructed using projection data normalized by the flood field image acquired inside the magnet, but not corrected for Lorentz shift is more accurate, but still distorted (Figure 2c). In contrast, the SPECT image reconstructed using projection data normalized by the flood field image acquired inside the magnet and also corrected for Lorentz shift is the most accurate of the three images (Figure 2d). These results demonstrate that uniformity correction of data acquired within an MRI system must be performed using a flood field image also acquired within the MRI at the appropriate location. Furthermore, data must be properly corrected for the Lorentz shift prior to tomographic reconstruction.

AC of the projection data does not appear to drastically improve the SPECT reconstruction seen in Figure 2e. However, close inspection of the profiles in Figure 3 reveals that AC slightly sharpens the phantom boundary and reduces signal loss in the central region of the reconstructed SPECT image. While these effects are subtle, their significance increases when imaging larger objects (e.g. humans) that generate greater attenuation. Never the less, even for these small-animal scales, gamma ray attenuation produces a measurable effect that degrades the accuracy of SPECT imaging. These results demonstrate the feasibility of utilizing simultaneously-acquired co-registered MR images for AC in SPECT.

For this study, we utilized a simple AC method appropriate for our (two-component) phantom where the ^99mTc radioisotope was known to be uniformly distributed within the water region. For the more general case of an unknown radioisotope distribution, we could still perform AC by first generating the SPECT reconstruction without AC, then using the resulting image (of the reconstructed radioisotope distribution) to perform the previously detailed AC. The corrected projection data could then be used to generate a more accurate SPECT reconstruction, and this process could then be iterated to further improve the results. Alternatively, other reported AC techniques could be applied to reconstruct the SPECT images of more complex objects (10).

Inspection of Figures 2f and 3 reveals that the accuracy of a lower spatial resolution SPECT image can be improved through the use of a corresponding higher spatial resolution MR image. In this specific case, the maximization of mutual information between the SPECT and MR images successfully optimized the Wiener filter utilized for image restoration. This represents a ‘post-processing’ algorithm, where the MRI data was utilized to improve the SPECT image after tomographic reconstruction of the projection data. Algorithms that integrate MRI information directly into the SPECT image reconstruction may offer improved performance and should be investigated further.

Animal Experiment

The axial MR magnitude image of the nude mouse is shown in Figure 4a. The corresponding reconstructed SPECT image is shown in Figure 4b. By itself, the exact location of the high activity region within the animal cannot be established due to a lack of anatomical detail. However, when the SPECT and MR images are fused together (Figure 4c), it becomes clearer that the high activity region corresponds to the ^99mTc-filled tube. These results demonstrate the utility of co-registered SPECT/MR imaging for more accurate localization of the radiotracers. For this preliminary study, the radioisotope was confined to a well defined region within the animal. For future studies, we plan to inject an appropriate radioisotope directly into a tumor-bearing animal to observe uptake of the tracer by the tumor.

Figure 4.

(a) Axial MR image of a nude mouse containing a tube filled with ^99mTc. (b) The co-registered reconstructed SPECT image. (c) Fusion of the SPECT and MR images for improved localization.

Additional Future Work

In addition to the development of integrated SPECT and MRI reconstruction algorithms and more detailed animal experiments, our MRSPECT system may be expanded for future studies. In this preliminary study, the nuclear radiation detector remained fixed while the object being imaged was manually rotated to obtain data from multiple views. Construction of a larger rotating gantry would allow for rotation of the detector about a (fixed) object. Furthermore, rotation of either the object or the detector could be automated through the use of an MR-compatible motor (11).

For micro-SPECT applications using pinhole collimators, our birdcage RF coil may be appropriately modified by cutting out regions for the pinholes. Likewise, our MRSPECT system may be utilized for human imaging through the suitable scaling of components. For a scaled up parallel-hole collimator larger than the space between the rungs of a birdcage coil, the collimator may simply be placed outside the coil and attenuation of the gamma rays by the coil materials taken into consideration during data processing. Alternatively, an array of receiver coil loops may be utilized instead of a birdcage coil, where a much larger space between the coil elements is permissible for placement of a collimator. Use of an RF receiver array with integrated nuclear radiation detectors can be implemented for both small-animal and human-sized scales with improved MR image quality (12). Since both SPECT and MRI modalities are applicable to both small animals and humans, we anticipate that MRSPECT will play a significant role in molecular imaging for various human diseases in the near future.

Abbreviations

SPECT

single-photon emission computed tomography

PET

positron emission tomography

MRI

magnetic resonance imaging

MRSPECT

combined SPECT and MRI, CZT, cadmium-zinc-telluride

ASIC

application-specific integrated circuit

RF

radiofrequency

PHS

pulse height spectrum

SNR

signal-to-noise ratio

SE

spin-echo

TR

repetition time

TE

echo time

FOV

field-of-view

NEX

number of excitations

AC

attenuation correction

FBP

filtered back-projection

FWHM

full width at half maximum