Implementation of GPU accelerated SPECT reconstruction with Monte Carlo-based scatter correction

Objective

Statistical SPECT reconstruction can be very time-consuming especially when compensations for collimator and detector response, attenuation, and scatter are included in the reconstruction. This work proposes an accelerated SPECT reconstruction algorithm based on graphics processing unit (GPU) processing.

Methods

Ordered subset expectation maximization (OSEM) algorithm with CT-based attenuation modelling, depth-dependent Gaussian convolution-based collimator-detector response modelling, and Monte Carlo-based scatter compensation was implemented using OpenCL. The OpenCL implementation was compared against the existing multi-threaded OSEM implementation running on a central processing unit (CPU) in terms of scatter-to-primary ratios, standardized uptake values (SUVs), and processing speed using mathematical phantoms and clinical multi-bed bone SPECT/CT studies.

Results

The difference in scatter-to-primary ratios, visual appearance, and SUVs between GPU and CPU implementations was minor. On the other hand, at its best, the GPU implementation was noticed to be 24 times faster than the multi-threaded CPU version on a normal 128?×?128 matrix size 3 bed bone SPECT/CT data set when compensations for collimator and detector response, attenuation, and scatter were included.

Conclusions

GPU SPECT reconstructions show great promise as an every day clinical reconstruction tool.

     

Scatter and attenuation correction in SPECT using density maps and Monte Carlo simulated scatter functions

A new scatter and attenuation correction method is presented in which Monte Carlo simulated scatter line-spread functions for different depth and lateral positions are used. A reconstructed emission image is used as an estimate of the source distribution in order to calculate the scatter contribution in the projection data. The scatter contribution is then subtracted from the original projection prior to attenuation correction. The attenuation correction method uses density maps for the attenuation correction of projection data. Simulation studies have been done with a clinically realistic source distribution in cylindrical, homogeneous water phantoms of different sizes and with photon energies corresponding to 201T1, 99mTc, and 111In. The results show excellent quantitative results with an accuracy within +/- 10% for most of the source positions and phantom sizes. It has also been shown that the variation in the event distribution within the source region in the images has been significantly decreased and that an enhancement in the contrast has been achieved.     

SPECT quantification: a simplified method of attenuation and scatter correction for cardiac imaging.

The quantitative and visual interpretation of SPECT myocardial perfusion images is limited by physical factors such as photon attenuation, Compton scatter, and finite resolution effects. A method of attenuation correction is described for use in nonhomogeneous media and applied to cardiac SPECT imaging. This method, termed multiplicative variable attenuation compensation (MVAC), uses tissue contours determined from segmentation of a transmission scan to assign a priori determined attenuation coefficients to different tissue regions of the transaxial images. An attenuation correction map is then constructed using a technique inspired by Chang's method that includes regionally dependent attenuation within the chest cavity and is applied after reconstruction by filtered backprojection. Scatter correction using the subtraction of a simultaneously acquired scatter window image enables the use of narrow beam attenuation coefficients. Experimental measurements to evaluate these methods were conducted for 201Tl and 99mTc SPECT using a homomorphic cardiac phantom. Finite resolution effects were included in the evaluation of results by computer simulation of the three-dimensional activity distribution. The correction methodology was shown to substantially improve both relative and absolute quantification of uniform and nonuniform regions of activity in the phantom's myocardial wall.     

Generalized scatter correction method in SPECT using point scatter distribution functions

A new two-dimensional (2-D) scatter correction technique in single photon emission computed tomography (SPECT) based on convolution or frequency filtering with a 2-D scatter distribution function is described. A scatter distribution function of the form A exp(-Br), has been derived from measurements of a point source in a water phantom. Both the amplitude A and the slope B of this function, were approximately invariant with source position except near phantom surface. The accuracy of the 2-D correction technique was compared with that of the previous one-dimensional (1-D) scatter correction technique. As could be expected the latter technique was shown to be less accurate due to its dependence on axial distribution of radioactivity. Phantom SPECT studies showed a clear superiority of the 2-D over the 1-D scatter correction in quantitative imaging. Images derived from clinical studies of regional bloodflow with 99mTc-HM-PAO and liver uptake showed significant contrast improvement by both techniques.     

Simultaneous 3-dimensional resolution correction in SPECT reconstruction with an ordered-subsets expectation maximization algorithm

Collimators are used for the improvement of information about the positions of sources by limiting the incidence direction of gamma-rays and characteristic x-rays to detectors. In this study, we attempted to improve the spatial resolution of (201)Tl myocardial SPECT by using simultaneous 3-dimensional distance-dependent resolution correction (DRC) incorporated into the ordered-subsets expectation maximization algorithm. METHODS: Simulation was performed with various sizes of balls, and measurement with a line-source phantom was performed at different source-detector distances. Imaging of a hot-rod phantom, the defect area of a myocardial phantom, and the myocardial thickness and cardiac lumen in a human study by (201)TlCl myocardial SPECT was evaluated before and after DRC. RESULTS: We performed simulation by using 5 sizes of balls and found marked improvement in resolution in all x-, y-, and z-axis directions. In the line-source phantom, when the radial distance was very long (30.5 cm), the correction effects were slightly low. However, when the distance was similar to the clinically used rotation radius (22.5 cm), the correction effects were good in the hot-rod and myocardial phantoms and in the human study. CONCLUSION: DRC markedly improved the spatial resolution of SPECT images, suggesting that this method is useful for the quantification of infarcted areas by myocardial SPECT.     

Estimation of scatter component in SPECT planar image using a Monte Carlo method

A Monte Carlo simulation was performed to estimate quantitatively the scattered photons in planar images of SPECT. As a phantom we used a water-filled cylinder with a line source and calculated the energy spectra of primary and multiple scattered photons separately at each pixel of the planar images. The energy spectra of primary and scattered photons were studied on following parameters: the size of the phantom; the location of the source; the width of the energy window centered at 72.3 (Tl-201), 141 (Tc-99m), and 159 keV (I-123); and the view angle of the planar images. Obtained results were; (1) the energy spectra of Compton scattered photon varied with the phantom size, the source location, and the photopeak energy, (2) the scattered photons within energy windows of 10-30% centered at the photopeak energy were mainly composed by Compton scatter of the first order, (3) the higher order scattering component of the Compton photons did not represent the location of the line source on the planar image, and (4) the scatter fraction defined by the ratio of the scattered photons to the primary photons increased with increasing the size of the phantom and the width of energy window at the low photopeak energy. From the results, we discussed on the scatter subtraction methods.     

An evaluation of reconstruction techniques and scatter correction in bone SPECT of the spine

A selection of commonly used reconstruction and filter techniques in the processing of 99mTc oxidronate (i.e., 99mTc hydroxymethane diphosphonate) single photon emission computed tomography (SPECT) of the spine was compared. The possible additional value of scatter correction on image contrast was also evaluated. Twenty-eight bone SPECT examinations of consecutive patients were studied retrospectively. The reconstruction techniques used were filtered back-projection and iterative reconstruction with the use of ordered subsets estimation maximization. Three-dimensional post-filtering with a Metz filter and a Butterworth filter was used. Each combination was evaluated with or without scatter correction. Each study was also processed with the department's standard technique of two-dimensional pre-filtering with a Metz filter followed by filtered back-projection (without scatter correction). Five observers evaluated the image quality of reconstructed coronal and sagittal slices, with special reference to the resolution of vertebrae, vertebral processes, the spinal canal and suspected abnormal uptakes. A grading scale from -2 to +2 was used with the standard technique as the reference. The best image quality was found with iterative reconstruction in combination with a contrast enhancing Metz filter or a noise reducing Butterworth filter. Scatter correction did not improve image quality.     

Absolute quantification of myocardial blood flow with 13N-ammonia and 3-dimensional PET.

The aim of this study was to compare 2-dimensional (2D) and 3-dimensional (3D) dynamic PET for the absolute quantification of myocardial blood flow (MBF) with (13)N-ammonia ((13)N-NH(3)). METHODS: 2D and 3D MBF measurements were collected from 21 patients undergoing cardiac evaluation at rest (n = 14) and during standard adenosine stress (n = 7). A lutetium yttrium oxyorthosilicate-based PET/CT system with retractable septa, enabling the sequential acquisition of 2D and 3D images within the same patient and study, was used. All 2D studies were performed by injecting 700-900 MBq of (13)N-NH(3). For 14 patients, 3D studies were performed with the same injected (13)N-NH(3) dose as that used in 2D studies. For the remaining 7 patients, 3D images were acquired with a lower dose of (13)N-NH(3), that is, 500 MBq. 2D images reconstructed by use of filtered backprojection (FBP) provided the reference standard for MBF measurements. 3D images were reconstructed by use of Fourier rebinning (FORE) with FBP (FORE-FBP), FORE with ordered-subsets expectation maximization (FORE-OSEM), and a reprojection algorithm (RP). RESULTS: Global MBF measurements derived from 3D PET with FORE-FBP (r = 0.97), FORE-OSEM (r = 0.97), and RP (r = 0.97) were well correlated with those derived from 2D FBP (all Ps < 0.0001). The mean +/- SD differences in global MBF measurements between 3D FORE-FBP and 2D FBP and between 3D FORE-OSEM and 2D FBP were 0.01 +/- 0.14 and 0.01 +/- 0.15 mL/min/g, respectively. The mean +/- SD difference in global MBF measurements between 3D RP and 2D FBP was 0.00 +/- 0.16 mL/min/g. The best correlation between 2D PET and 3D PET performed with the lower injected activity was found for the 3D FORE-FBP reconstruction algorithm (r = 0.95, P < 0.001). CONCLUSION: For this scanner type, quantitative measurements of MBF with 3D PET and (13)N-NH(3) were in excellent agreement with those obtained with the 2D technique, even when a lower activity was injected.     

Evaluation of 2 scatter correction methods using a striatal phantom for quantitative brain SPECT

OBJECTIVE: Scatter correction is an important factor in quantitative SPECT. In this study, we evaluated 2 methods of scatter correction for brain SPECT. The first is based on thresholding the energy spectrum (ES), and the second is based on a modification of the transmission-dependent convolution subtraction (TDCS) method. METHODS: SPECT imaging of a skull striatal phantom was performed using a triple-head camera with and without scatter correction. The striatal compartments were filled with (123)I, and the brain shell cavity (background) was filled with varying concentrations of (123)I to obtain striatal-to-background ratios of 2, 5, 10, 15, 20, and 25 to 1, respectively, which were considered to be the expected ratios. SPECT-measured ratios of striatal-to-background counts were determined with scatter correction (both ES and TDCS methods) and without scatter correction and were then compared with the expected ratios. RESULTS: Without scatter correction, measured striatal-to-background ratios were underestimated by an average of 41.7%, compared with the expected ratios. The ES method of scatter correction underestimated the striatal-to-background ratios by an average of 27.4%, a significant improvement (P < 0.04) over those without scatter correction. With the TDCS method of scatter correction, the ratios were underestimated by only 3.3% (P < 0.03). TDCS ratios were significantly (P < 0.04) higher than ES ratios and were nearly identical to the expected ratios. CONCLUSION: These results suggest that scatter correction significantly improves the striatal-to-background ratios. The TDCS method appears to correct scatter more effectively than does the ES method for the striatal phantom, thus providing more accurate quantification.     

Use of a germanium detector to optimize scatter correction in SPECT

A collimated germanium detector with an energy resolution of 1 keV full width at half maximum at 140 keV was used to measure the energy spectrum of radiation emitted from a test object containing an asymmetric distribution of 99mTc and nonuniform attenuation. Energy spectra were recorded from 24 positions around the object and convolved with Gaussian functions to simulate data that would have been acquired with a scintillation camera. The scatter fraction was computed from the convolved spectra in conjunction with a scatter-free reference spectrum. After adding appropriate Poisson noise, a technique based on maximizing the signal to noise ratio was developed to optimally subtract the scatter fraction from the recorded counts. SPECT imaging of the test object was performed to evaluate the correction technique.     

Quantification of dopaminergic neurotransmission SPECT studies with 123I-labelled radioligands. A comparison between different imaging systems and data acquisition protocols using Monte Carlo simulation

PURPOSE: (123)I-labelled radioligands are commonly used for single-photon emission computed tomography (SPECT) imaging of the dopaminergic system to study the dopamine transporter binding. The aim of this work was to compare the quantitative capabilities of two different SPECT systems through Monte Carlo (MC) simulation. METHODS: The SimSET MC code was employed to generate simulated projections of a numerical phantom for two gamma cameras equipped with a parallel and a fan-beam collimator, respectively. A fully 3D iterative reconstruction algorithm was used to compensate for attenuation, the spatially variant point spread function (PSF) and scatter. A post-reconstruction partial volume effect (PVE) compensation was also developed. RESULTS: For both systems, the correction for all degradations and PVE compensation resulted in recovery factors of the theoretical specific uptake ratio (SUR) close to 100%. For a SUR value of 4, the recovered SUR for the parallel imaging system was 33% for a reconstruction without corrections (OSEM), 45% for a reconstruction with attenuation correction (OSEM-A), 56% for a 3D reconstruction with attenuation and PSF corrections (OSEM-AP), 68% for OSEM-AP with scatter correction (OSEM-APS) and 97% for OSEM-APS plus PVE compensation (OSEM-APSV). For the fan-beam imaging system, the recovered SUR was 41% without corrections, 55% for OSEM-A, 65% for OSEM-AP, 75% for OSEM-APS and 102% for OSEM-APSV. CONCLUSION: Our findings indicate that the correction for degradations increases the quantification accuracy, with PVE compensation playing a major role in the SUR quantification. The proposed methodology allows us to reach similar SUR values for different SPECT systems, thereby allowing a reliable standardisation in multicentric studies.     

Optimization of iterative reconstruction parameters with attenuation correction,scatter correction and resolution recovery in myocardial perfusion SPECT/CT

Objective

The aim of this study was to characterize the optimal reconstruction parameters for ordered-subset expectation maximization (OSEM) with attenuation correction, scatter correction, and depth-dependent resolution recovery (OSEM_ACSCRR). We assessed the optimal parameters for OSEM_ACSCRR in an anthropomorphic torso phantom study, and evaluated the validity of the reconstruction parameters in the groups of normal volunteers and patients with abnormal perfusion.

Methods

Images of the anthropomorphic torso phantom, 9 normal volunteers and 7 patients undergoing myocardial perfusion SPECT were acquired with a SPECT/CT scanner. SPECT data comprised a 64 × 64 matrix with an acquisition pixel size of 6.6 mm. A normalized mean square error (NMSE) of the phantom image was calculated to determine both optimal OSEM update and a full width at half maximum (FWHM) of Gaussian filter. We validated the myocardial count, contrast and noise characteristic for clinical subjects derived from OSEM_ACSCRR processing. OSEM with depth-dependent resolution recovery (OSEM_RR) and filtered back projection (FBP) were simultaneously performed to compare OSEM_ACSCRR.

Results

The combination of OSEM_ACSCRR with 90–120 OSEM updates and Gaussian filter with 13.2–14.85 mm FWHM yielded low NMSE value in the phantom study. When we used OSEM_ACSCRR with 120 updates and Gaussian filter with 13.2 mm FWHM in the normal volunteers, myocardial contrast showed significantly higher value than that derived from 120 updates and 14.85 mm FWHM. OSEM_ACSCRR with the combination of 90–120 OSEM updates and 14.85 mm FWHM produced lowest % root mean square (RMS) noise. Regarding the defect contrast of patients with abnormal perfusion, OSEM_ACSCRR with the combination of 90–120 OSEM updates and 13.2 mm FWHM produced significantly higher value than that derived from 90–120 OSEM updates and 14.85 mm FWHM. OSEM_ACSCRR was superior to FBP for the % RMS noise (8.52 ± 1.08 vs. 9.55 ± 1.71, p = 0.02) and defect contrast (0.368 ± 0.061 vs. 0.327 ± 0.052, p = 0.01), respectively.

Conclusions

Clinically optimized the number of OSEM updates and FWHM of Gaussian filter were (1) 120 updates and 13.2 mm, and (2) 90–120 updates and 14.85 mm on the OSEM_ACSCRR processing, respectively. Further assessment may be required to determine the optimal iterative reconstruction parameters in a larger patient population.     

A technique for using CT images in attenuation correction and quantification in SPECT

A technique is described for using computed tomography (CT) images for attenuation correction and quantification in SPECT. The CT images are aligned with the corresponding SPECT slices and the Hounsfield units are converted to linear attenuation coefficient values for the SPECT radionuclide. The attenuation coefficient map thus produced is used to provide the attenuation correction required in the SPECT reconstruction. The technique has been evaluated in both a non-anatomical and an anatomical phantom giving a mean accuracy in quantifying activity of various features in the phantoms of 2.6% (range 0.3%-4.0%). The value of performing scatter correction prior to attenuation correction in obtaining accurate quantification is demonstrated. The practicalities of applying the technique in patient studies are discussed.     

Improvement of quantification in SPECT studies by scatter and attenuation compensation

The filtered backprojection images obtained from classical SPECT studies are not adequate for evaluation of volumes or parameters of clinical interest. Noise, scattering, boundary accuracy and attenuation are the main problems of SPECT quantification. It is the aim of the following study to overcome these difficulties. The first step of all correction algorithm is the contour detection of the attenuation medium. A new procedure, previously described by the authors, accurately and automatically found the boundaries of the surrounding body. The Compton scattering climination is carried out by a modified version of Jaszczak's method. This alteration is essential to implement the iterative attenuation correction algorithm derived from Chang's method. Results obtained using computer simulation and real phantoms or clinical studies demonstrate the high improvement of contrast and count levels in the corrected slices. The process is fully automatic and the efficiency of the procedures allow fast processing of the daily SPECT examination.     

Improvement of quantification in SPECT studies by scatter and attenuation compensation

The filtered backprojection images obtained from classical SPECT studies are not adequate for evaluation of volumes or parameters of clinical interest. Noise, scattering, boundary accuracy and attenuation are the main problems of SPECT quantification. It is the aim of the following study to overcome these difficulties. The first step of all correction algorithm is the contour detection of the attenuation medium. A new procedure, previously described by the authors, accurately and automatically found the boundaries of the surrounding body. The Compton scattering elimination is carried out by a modified version of Jaszczak's method. This alteration is essential to implement the iterative attenuation correction algorithm derived from Chang's method. Results obtained using computer simulation and real phantoms or clinical studies demonstrate the high improvement of contrast and count levels in the corrected slices. The process is fully automatic and the efficiency of the procedures allow fast processing of the daily SPECT examination.     

Improved quantification in multiple-pinhole SPECT by anatomy-based reconstruction using microCT information

Purpose  

The aim of this study was to evaluate the potential of anatomy-based reconstruction, using microCT information, to improve quantitative accuracy in multiple-pinhole SPECT.     

Impaired dopaminergic neurotransmission in patients with traumatic brain injury: a SPECT study using 123I-beta-CIT and 123I-IBZM

Structural imaging suggests that traumatic brain injury (TBI) may be associated with disruption of neuronal networks, including the nigrostriatal dopaminergic pathway. However, to date deficits in pre- and/or postsynaptic dopaminergic neurotransmission have not been demonstrated in TBI using functional imaging. We therefore assessed dopaminergic function in ten TBI patients using [123I]2-beta-carbomethoxy-3-beta-(4-iodophenyl)tropane (beta-CIT) and [123I]iodobenzamide (IBZM) single-photon emission tomography (SPET). Average Glasgow Coma Scale score (+/-SD) at the time of head trauma was 5.8+/-4.2. SPET was performed on average 141 days (SD +/-92) after TBI. The SPET images were compared with structural images using cranial computerised tomography (CCT) and magnetic resonance imaging (MRI). SPET was performed with an ADAC Vertex dual-head camera. The activity ratios of striatal to cerebellar uptake were used as a semiquantitative parameter of striatal dopamine transporter (DAT) and D2 receptor (D2R) binding. Compared with age-matched controls, patients with TBI had significantly lower striatal/cerebellar beta-CIT and IBZM binding ratios (P< or =0.01). Overall, the DAT deficit was more marked than the D2R loss. CCT and MRI studies revealed varying cortical and subcortical lesions, with the frontal lobe being most frequently affected whereas the striatum appeared structurally normal in all but one patient. Our findings suggest that nigrostriatal dysfunction may be detected using SPET following TBI despite relative structural preservation of the striatum. Further investigations of possible clinical correlates and efficacy of dopaminergic therapy in patients with TBI seem justified.     

Introducing simultaneous spatial resolution and attenuation correction after scatter removal in SPECT imaging.

A new approach to simultaneous spatial resolution and attenuation correction in SPECT imaging is presented. Before these corrections, scatter is removed on the projections. This removal is performed by spectral constrained factor analysis. The innovation reported here is the use of the different impulse responses of the system, according to the source-detector distance, and their integration in a generalized version of the Chang attenuation correction method. This novel algorithm is evaluated on computed and physical phantoms. In the computer-simulated phantom, the count rates after full-processing are very close to the initial values. In the physical phantom, the contrast is increased by 1.8 after full processing. The activity profiles drawn both on raw projections and reconstructed slices demonstrate the effectiveness of the algorithm for the restoration of spatial resolution. Furthermore, the method improves the quality of the images greatly. A clinical study is also presented. When the whole procedure is applied, the resulting slice matches the corresponding computed tomographic scan very well, which is not the case with the usual back-projected images. The process is fully automatic and the computing time performance allows its daily use for single photon emission tomographic examinations.     

Accurate dosimetry in 131I radionuclide therapy using patient-specific, 3-dimensional methods for SPECT reconstruction and absorbed dose calculation.

(131)I radionuclide therapy studies have not shown a strong relationship between tumor absorbed dose and response, possibly due to inaccuracies in activity quantification and dose estimation. The goal of this work was to establish the accuracy of (131)I activity quantification and absorbed dose estimation when patient-specific, 3-dimensional (3D) methods are used for SPECT reconstruction and for absorbed dose calculation. METHODS: Clinically realistic voxel-phantom simulations were used in the evaluation of activity quantification and dosimetry. SPECT reconstruction was performed using an ordered-subsets expectation maximization (OSEM) algorithm with compensation for scatter, attenuation, and 3D detector response. Based on the SPECT image and a patient-specific density map derived from CT, 3D dosimetry was performed using a newly implemented Monte Carlo code. Dosimetry was evaluated by comparing mean absorbed dose estimates calculated directly from the defined phantom activity map with those calculated from the SPECT image of the phantom. Finally, the 3D methods were applied to a radioimmunotherapy patient, and the mean tumor absorbed dose from the new calculation was compared with that from conventional dosimetry obtained from conjugate-view imaging. RESULTS: Overall, the accuracy of the SPECT-based absorbed dose estimates in the phantom was >12% for targets down to 16 mL and up to 35% for the smallest 7-mL tumor. To improve accuracy in the smallest tumor, more OSEM iterations may be needed. The relative SD from multiple realizations was <3% for all targets except for the smallest tumor. For the patient, the mean tumor absorbed dose estimate from the new Monte Carlo calculation was 7% higher than that from conventional dosimetry. CONCLUSION: For target sizes down to 16 mL, highly accurate and precise dosimetry can be obtained with 3D methods for SPECT reconstruction and absorbed dose estimation. In the future, these methods can be applied to patients to potentially establish correlations between tumor regression and the absorbed dose statistics from 3D dosimetry.