Simon Stute

GATE V6: a major enhancement of the GATE simulation platform enabling modelling of CT and radiotherapy

GATE (Geant4 Application for Emission Tomography) is a Monte Carlo simulation platform developed by the OpenGATE collaboration since 2001 and first publicly released in 2004. Dedicated to the modelling of planar scintigraphy, single photon emission computed tomography (SPECT) and positron emission tomography (PET) acquisitions, this platform is widely used to assist PET and SPECT research. A recent extension of this platform, released by the OpenGATE collaboration as GATE V6, now also enables modelling of x-ray computed tomography and radiation therapy experiments. This paper presents an overview of the main additions and improvements implemented in GATE since the publication of the initial GATE paper (Jan et al 2004 Phys. Med. Biol. 49 4543-61). This includes new models available in GATE to simulate optical and hadronic processes, novelties in modelling tracer, organ or detector motion, new options for speeding up GATE simulations, examples illustrating the use of GATE V6 in radiotherapy applications and CT simulations, and preliminary results regarding the validation of GATE V6 for radiation therapy applications. Upon completion of extensive validation studies, GATE is expected to become a valuable tool for simulations involving both radiotherapy and imaging.

Comparative assessment of methods for estimating tumor volume and standardized uptake value in (18)F-FDG PET.

In 18F-FDG PET, tumors are often characterized by their metabolically active volume and standardized uptake value (SUV). However, many approaches have been proposed to estimate tumor volume and SUV from 18F-FDG PET images, none of them being widely agreed upon. We assessed the accuracy and robustness of 5 methods for tumor volume estimates and of 10 methods for SUV estimates in a large variety of configurations. Methods: PET acquisitions of an anthropomorphic phantom containing 17 spheres (volumes between 0.43 and 97 mL, sphere-to-surrounding-activity concentration ratios between 2 and 68) were used. Forty-one nonspheric tumors (volumes between 0.6 and 92 mL, SUV of 2, 4, and 8) were also simulated and inserted in a real patient 18F-FDG PET scan. Four threshold-based methods (including one, Tbgd, accounting for background activity) and a model-based method (Fit) described in the literature were used for tumor volume measurements. The mean SUV in the resulting volumes were calculated, without and with partial-volume effect (PVE) correction, as well as the maximum SUV (SUVmax). The parameters involved in the tumor segmentation and SUV estimation methods were optimized using 3 approaches, corresponding to getting the best of each method or testing each method in more realistic situations in which the parameters cannot be perfectly optimized. Results: In the phantom and simulated data, the Tbgd and Fit methods yielded the most accurate volume estimates, with mean errors of 2% ± 11% and −8% ± 21% in the most realistic situations. Considering the simulated data, all SUV not corrected for PVE had a mean bias between −31% and −46%, much larger than the bias observed with SUVmax (−11% ± 23%) or with the PVE-corrected SUV based on Tbgd and Fit (−2% ± 10% and 3% ± 24%). Conclusion: The method used to estimate tumor volume and SUV greatly affects the reliability of the estimates. The Tbgd and Fit methods yielded low errors in volume estimates in a broad range of situations. The PVE-corrected SUV based on Tbgd and Fit were more accurate and reproducible than SUVmax.

A New Method for Volume Segmentation of PET Images, Based on Possibility Theory

18F-fluorodeoxyglucose positron emission tomography (18FDG PET) has become an essential technique in oncology. Accurate segmentation and uptake quantification are crucial in order to enable objective follow-up, the optimization of radiotherapy planning, and therapeutic evaluation. We have designed and evaluated a new, nearly automatic and operator-independent segmentation approach. This incorporated possibility theory, in order to take into account the uncertainty and inaccuracy inherent in the image. The approach remained independent of PET facilities since it did not require any preliminary calibration. Good results were obtained from phantom images [percent error =18.38% (mean) ±9.72% (standard deviation)]. Results on simulated and anatomopathological data sets were quantified using different similarity measures and showed the method was efficient (simulated images: Dice index =82.18% ±13.53% for SUV =2.5 ). The approach could, therefore, be an efficient and robust tool for uptake volume segmentation, and lead to new indicators for measuring volume of interest activity.

Fully 4D list-mode reconstruction applied to respiratory-gated PET scans

(18)F-fluoro-deoxy-glucose ((18)F-FDG) positron emission tomography (PET) is one of the most sensitive and specific imaging modalities for the diagnosis of non-small cell lung cancer. A drawback of PET is that it requires several minutes of acquisition per bed position, which results in images being affected by respiratory blur. Respiratory gating techniques have been developed to deal with respiratory motion in the PET images. However, these techniques considerably increase the level of noise in the reconstructed images unless the acquisition time is increased. The aim of this paper is to evaluate a four-dimensional (4D) image reconstruction algorithm that combines the acquired events in all the gates whilst preserving the motion deblurring. This algorithm was compared to classic ordered subset expectation maximization (OSEM) reconstruction of gated and non-gated images, and to temporal filtering of gated images reconstructed with OSEM. Two datasets were used for comparing the different reconstruction approaches: one involving the NEMA IEC/2001 body phantom in motion, the other obtained using Monte-Carlo simulations of the NCAT breathing phantom. Results show that 4D reconstruction reaches a similar performance in terms of the signal-to-noise ratio (SNR) as non-gated reconstruction whilst preserving the motion deblurring. In particular, 4D reconstruction improves the SNR compared to respiratory-gated images reconstructed with the OSEM algorithm. Temporal filtering of the OSEM-reconstructed images helps improve the SNR, but does not achieve the same performance as 4D reconstruction. 4D reconstruction of respiratory-gated images thus appears as a promising tool to reach the same performance in terms of the SNR as non-gated acquisitions while reducing the motion blur, without increasing the acquisition time.

Practical considerations for image-based PSF and blobs reconstruction in PET

Iterative reconstructions in positron emission tomography (PET) need a model relating the recorded data to the object/patient being imaged, called the system matrix (SM). The more realistic this model, the better the spatial resolution in the reconstructed images. However, a serious concern when using a SM that accurately models the resolution properties of the PET system is the undesirable edge artefact, visible through oscillations near sharp discontinuities in the reconstructed images. This artefact is a natural consequence of solving an ill-conditioned inverse problem, where the recorded data are band-limited. In this paper, we focus on practical aspects when considering image-based point-spread function (PSF) reconstructions. To remove the edge artefact, we propose to use a particular case of the method of sieves (Grenander 1981 Abstract Inference New York: Wiley), which simply consists in performing a standard PSF reconstruction, followed by a post-smoothing using the PSF as the convolution kernel. Using analytical simulations, we investigate the impact of different reconstruction and PSF modelling parameters on the edge artefact and its suppression, in the case of noise-free data and an exactly known PSF. Using Monte-Carlo simulations, we assess the proposed method of sieves with respect to the choice of the geometric projector and the PSF model used in the reconstruction. When the PSF model is accurately known, we show that the proposed method of sieves succeeds in completely suppressing the edge artefact, though after a number of iterations higher than typically used in practice. When applying the method to realistic data (i.e. unknown true SM and noisy data), we show that the choice of the geometric projector and the PSF model does not impact the results in terms of noise and contrast recovery, as long as the PSF has a width close to the true PSF one. Equivalent results were obtained using either blobs or voxels in the same conditions (i.e. the blobs density function being the same as the voxel-based PSF). From a practical point-of-view, the method can be used to perform fast reconstructions based on very simple models (compared to sinogram-based PSF modelling), producing artefact-free images with a better compromise between noise and spatial resolution than images reconstructed without or with under-estimated PSF. Besides, the method inherently limits the spatial resolution in the reconstructed images to the intrinsic one of the PET system.

Monte Carlo simulations of clinical PET and SPECT scans: impact of the input data on the simulated images

Monte Carlo simulations of emission tomography have proven useful to assist detector design and optimize acquisition and processing protocols. The more realistic the simulations, the more straightforward the extrapolation of conclusions to clinical situations. In emission tomography, accurate numerical models of tomographs have been described and well validated under specific operating conditions (collimator, radionuclide, acquisition parameters, count rates, etc). When using these models under these operating conditions, the realism of simulations mostly depends on the activity distribution used as an input for the simulations. It has been proposed to derive the input activity distribution directly from reconstructed clinical images, so as to properly model the heterogeneity of the activity distribution between and within organs. However, reconstructed patient images include noise and have limited spatial resolution. In this study, we analyse the properties of the simulated images as a function of the properties of the reconstructed images used to define the input activity distributions in (18)F-FDG PET and (131)I SPECT simulations. The propagation through the simulation/reconstruction process of the noise and spatial resolution in the input activity distribution was studied using simulations. We found that the noise properties of the images reconstructed from the simulated data were almost independent of the noise in the input activity distribution. The spatial resolution in the images reconstructed from the simulations was slightly poorer than that in the input activity distribution. However, using high-noise but high-resolution patient images as an input activity distribution yielded reconstructed images that could not be distinguished from clinical images. These findings were confirmed by simulated highly realistic (131)I SPECT and (18)F-FDG PET images from patient data. In conclusion, we demonstrated that (131)I SPECT and (18)F-FDG PET images indistinguishable from real scans can be simulated using activity maps with spatial resolution higher than that used in routine clinical applications.

A method for accurate modelling of the crystal response function at a crystal sub-level applied to PET reconstruction

Positron emission tomography (PET) images suffer from low spatial resolution and signal-to-noise ratio. Accurate modelling of the effects affecting resolution within iterative reconstruction algorithms can improve the trade-off between spatial resolution and signal-to-noise ratio in PET images. In this work, we present an original approach for modelling the resolution loss introduced by physical interactions between and within the crystals of the tomograph and we investigate the impact of such modelling on the quality of the reconstructed images. The proposed model includes two components: modelling of the inter-crystal scattering and penetration (interC) and modelling of the intra-crystal count distribution (intraC). The parameters of the model were obtained using a Monte Carlo simulation of the Philips GEMINI GXL response. Modelling was applied to the raw line-of-response geometric histograms along the four dimensions and introduced in an iterative reconstruction algorithm. The impact of modelling interC, intraC or combined interC and intraC on spatial resolution, contrast recovery and noise was studied using simulated phantoms. The feasibility of modelling interC and intraC in two clinical (18)F-NaF scans was also studied. Measurements on Monte Carlo simulated data showed that, without any crystal interaction modelling, the radial spatial resolution in air varied from 5.3 mm FWHM at the centre of the field-of-view (FOV) to 10 mm at 266 mm from the centre. Resolution was improved with interC modelling (from 4.4 mm in the centre to 9.6 mm at the edge), or with intraC modelling only (from 4.8 mm in the centre to 4.3 mm at the edge), and it became stationary across the FOV (4.2 mm FWHM) when combining interC and intraC modelling. This improvement in resolution yielded significant contrast enhancement, e.g. from 65 to 76% and 55.5 to 68% for a 6.35 mm radius sphere with a 3.5 sphere-to-background activity ratio at 55 and 215 mm from the centre of the FOV, respectively, without introducing additional noise. Patient images confirmed the usefulness of interC and intraC modelling for improving spatial resolution and contrast. Based on Monte Carlo simulated data, we conclude that four-dimensional modelling of the inter- and intra-crystal interactions during the reconstruction process yields a significantly improved contrast to noise ratio and the stationarity of the spatial resolution in the reconstructed images.

Realistic and Efficient Modeling of Radiotracer Heterogeneity in Monte Carlo Simulations of PET Images With Tumors

Monte Carlo simulations are extensively used in PET to evaluate the accuracy with which PET images can yield reliable estimates of parameters of interest. For such applications, the simulated images should be as realistic as possible so that conclusions can be extrapolated to clinical PET images. In this work, we describe a method for introducing realistic modeling of radiotracer heterogeneity into Monte Carlo simulations of patient PET scans. The modeling of the complex physiological activity distribution in healthy regions is directly based on real patient PET/CT images, and realistic tumor shapes can be included into these regions. This method represents a competitive alternative to the use of complex anthropomorphic phantoms such as the XCAT, that require a fixed activity per structure. The method is extended to the simulation of serial PET scans with tumor changes, as acquired in the context of therapy monitoring, and this extension is validated using a patient study. Using the proposed method, very realistic patient PET images can be produced for evaluation purposes.In addition, a strategy to efficiently simulate many sets of pathological cases, based on a unique background physiological activity distribution, is described and carefully assessed using a numerical phantom. The background activity is simulated only once, while tumors are simulated separately. The data are then recombined in a specific way so that the final image has the same properties as images produced by simulating pathological and tumor activities at the same time.

Maximum-likelihood reconstruction based on a modified Poisson distribution to reduce bias in PET

The AB-ML algorithm [2] and the proposed NEG-ML algorithm can reduce or eliminate the bias in reconstructions from noisy data. AB-ML shifts the Poisson distribution, while NEG-ML modifies it near zero. Both algorithms have a single parameter that defines the balance between MLEM-like and unweighted-least-squares-like behaviour.

Bias Reduction for Low-Statistics PET: Maximum Likelihood Reconstruction With a Modified Poisson Distribution

Positron emission tomography data are typically reconstructed with maximum likelihood expectation maximization (MLEM). However, MLEM suffers from positive bias due to the non-negativity constraint. This is particularly problematic for tracer kinetic modeling. Two reconstruction methods with bias reduction properties that do not use strict Poisson optimization are presented and compared to each other, to filtered backprojection (FBP), and to MLEM. The first method is an extension of NEGML, where the Poisson distribution is replaced by a Gaussian distribution for low count data points. The transition point between the Gaussian and the Poisson regime is a parameter of the model. The second method is a simplification of ABML. ABML has a lower and upper bound for the reconstructed image whereas AML has the upper bound set to infinity. AML uses a negative lower bound to obtain bias reduction properties. Different choices of the lower bound are studied. The parameter of both algorithms determines the effectiveness of the bias reduction and should be chosen large enough to ensure bias-free images. This means that both algorithms become more similar to least squares algorithms, which turned out to be necessary to obtain bias-free reconstructions. This comes at the cost of increased variance. Nevertheless, NEGML and AML have lower variance than FBP. Furthermore, randoms handling has a large influence on the bias. Reconstruction with smoothed randoms results in lower bias compared to reconstruction with unsmoothed randoms or randoms precorrected data. However, NEGML and AML yield both bias-free images for large values of their parameter.