A unified approach to statistical tomography using coordinate descent optimization

Sparse representation and dictionary learning penalized image reconstruction for positron emission tomography

Accurate and robust reconstruction of the radioactivity concentration is of great importance in positron emission tomography (PET) imaging. Given the Poisson nature of photo-counting measurements, we present a reconstruction framework that integrates sparsity penalty on a dictionary into a maximum likelihood estimator. Patch-sparsity on a dictionary provides the regularization for our effort, and iterative procedures are used to solve the maximum likelihood function formulated on Poisson statistics. Specifically, in our formulation, a dictionary could be trained on CT images, to provide intrinsic anatomical structures for the reconstructed images, or adaptively learned from the noisy measurements of PET. Accuracy of the strategy with very promising application results from Monte-Carlo simulations, and real data are demonstrated.

Evaluation of penalty design in penalized maximum-likelihood image reconstruction for lesion detection

Detecting cancerous lesions is a major clinical application in emission tomography. Previously, we developed a method to design a shift-variant quadratic penalty function in penalized maximum-likelihood (PML) image reconstruction to improve the lesion detectability. We used a multiview channelized Hotelling observer (mvCHO) to assess the lesion detectability in three-dimensional images and validated the penalty design using computer simulations. In this study, we evaluate the benefit of the proposed penalty function for lesion detection using real patient data and artificial lesions. A high-count real patient dataset with no identifiable tumor inside the field of view is used as the background data. A Na-22 point source is scanned in air at variable locations and the point source data are superimposed onto the patient data as artificial lesions after being attenuated by the patient body. Independent Poisson noise is introduced to the high-count sinograms to generate 200 pairs of lesion-present and lesion-absent datasets, each mimicking a 5-min scan. Lesion detectability is assessed using a mvCHO and a human observer two-alternative forced choice (2AFC) experiment. The results show improvements in lesion detection by the proposed method compared with the conventional first-order quadratic penalty function and a total variation (TV) edge-preserving penalty function.

Statistical model based iterative reconstruction (MBIR) in clinical CT systems: experimental assessment of noise performance

PURPOSE

To reduce radiation dose in CT imaging, the statistical model based iterative reconstruction (MBIR) method has been introduced for clinical use. Based on the principle of MBIR and its nonlinear nature, the noise performance of MBIR is expected to be different from that of the well-understood filtered backprojection (FBP) reconstruction method. The purpose of this work is to experimentally assess the unique noise characteristics of MBIR using a state-of-the-art clinical CT system.

METHODS

Three physical phantoms, including a water cylinder and two pediatric head phantoms, were scanned in axial scanning mode using a 64-slice CT scanner (Discovery CT750 HD, GE Healthcare, Waukesha, WI) at seven different mAs levels (5, 12.5, 25, 50, 100, 200, 300). At each mAs level, each phantom was repeatedly scanned 50 times to generate an image ensemble for noise analysis. Both the FBP method with a standard kernel and the MBIR method (Veo(®), GE Healthcare, Waukesha, WI) were used for CT image reconstruction. Three-dimensional (3D) noise power spectrum (NPS), two-dimensional (2D) NPS, and zero-dimensional NPS (noise variance) were assessed both globally and locally. Noise magnitude, noise spatial correlation, noise spatial uniformity and their dose dependence were examined for the two reconstruction methods.

RESULTS

(1) At each dose level and at each frequency, the magnitude of the NPS of MBIR was smaller than that of FBP. (2) While the shape of the NPS of FBP was dose-independent, the shape of the NPS of MBIR was strongly dose-dependent; lower dose lead to a "redder" NPS with a lower mean frequency value. (3) The noise standard deviation (σ) of MBIR and dose were found to be related through a power law of σ ∝ (dose)(-β) with the component β ≈ 0.25, which violated the classical σ ∝ (dose)(-0.5) power law in FBP. (4) With MBIR, noise reduction was most prominent for thin image slices. (5) MBIR lead to better noise spatial uniformity when compared with FBP. (6) A composite image generated from two MBIR images acquired at two different dose levels (D1 and D2) demonstrated lower noise than that of an image acquired at a dose level of D1+D2.

CONCLUSIONS

The noise characteristics of the MBIR method are significantly different from those of the FBP method. The well known tradeoff relationship between CT image noise and radiation dose has been modified by MBIR to establish a more gradual dependence of noise on dose. Additionally, some other CT noise properties that had been well understood based on the linear system theory have also been altered by MBIR. Clinical CT scan protocols that had been optimized based on the classical CT noise properties need to be carefully re-evaluated for systems equipped with MBIR in order to maximize the method's potential clinical benefits in dose reduction and/or in CT image quality improvement.

Deriving adaptive MRF coefficients from previous normal-dose CT scan for low-dose image reconstruction via penalized weighted least-squares minimization

PURPOSE

Repeated computed tomography (CT) scans are required for some clinical applications such as image-guided interventions. To optimize radiation dose utility, a normal-dose scan is often first performed to set up reference, followed by a series of low-dose scans for intervention. One common strategy to achieve the low-dose scan is to lower the x-ray tube current and exposure time (mAs) or tube voltage (kVp) setting in the scanning protocol, but the resulted image quality by the conventional filtered back-projection (FBP) method may be severely degraded due to the excessive noise. Penalized weighted least-squares (PWLS) image reconstruction has shown the potential to significantly improve the image quality from low-mAs acquisitions, where the penalty plays an important role. In this work, the authors' explore an adaptive Markov random field (MRF)-based penalty term by utilizing previous normal-dose scan to improve the subsequent low-dose scans image reconstruction.

METHODS

In this work, the authors employ the widely-used quadratic-form MRF as the penalty model and explore a novel idea of using the previous normal-dose scan to obtain the MRF coefficients for adaptive reconstruction of the low-dose images. In the coefficients determination, the authors further explore another novel idea of using the normal-dose scan to obtain a scale map, which describes an optimal neighborhood for the coefficients determination such that a local uniform region has a small spread of frequency spectrum and, therefore, a small MRF window, and vice versa. The proposed penalty term is incorporated into the PWLS image reconstruction framework, and the low-dose images are reconstructed via the PWLS minimization.

RESULTS

The presented adaptive MRF based PWLS algorithm was validated by physical phantom and patient data. The experimental results demonstrated that the presented algorithm is superior to the PWLS reconstruction using the conventional Gaussian MRF penalty or the edge-preserving Huber penalty and the conventional FBP method, in terms of image noise reduction and edge/detail/contrast preservation.

CONCLUSIONS

This study demonstrated the feasibility and efficacy of the proposed scheme in utilizing previous normal-dose CT scan to improve the subsequent low-dose scans.

Statistical image reconstruction for low-dose CT using nonlocal means-based regularization

Low-dose computed tomography (CT) imaging without sacrifice of clinical tasks is desirable due to the growing concerns about excessive radiation exposure to the patients. One common strategy to achieve low-dose CT imaging is to lower the milliampere-second (mAs) setting in data scanning protocol. However, the reconstructed CT images by the conventional filtered back-projection (FBP) method from the low-mAs acquisitions may be severely degraded due to the excessive noise. Statistical image reconstruction (SIR) methods have shown potentials to significantly improve the reconstructed image quality from the low-mAs acquisitions, wherein the regularization plays a critical role and an established family of regularizations is based on the Markov random field (MRF) model. Inspired by the success of nonlocal means (NLM) in image processing applications, in this work, we propose to explore the NLM-based regularization for SIR to reconstruct low-dose CT images from low-mAs acquisitions. Experimental results with both digital and physical phantoms consistently demonstrated that SIR with the NLM-based regularization can achieve more gains than SIR with the well-known Gaussian MRF regularization or the generalized Gaussian MRF regularization and the conventional FBP method, in terms of image noise reduction and resolution preservation.

Statistical model based iterative reconstruction (MBIR) in clinical CT systems. Part II. Experimental assessment of spatial resolution performance

PURPOSE

Statistical model based iterative reconstruction (MBIR) methods have been introduced to clinical CT systems and are being used in some clinical diagnostic applications. The purpose of this paper is to experimentally assess the unique spatial resolution characteristics of this nonlinear reconstruction method and identify its potential impact on the detectabilities and the associated radiation dose levels for specific imaging tasks.

METHODS

The thoracic section of a pediatric phantom was repeatedly scanned 50 or 100 times using a 64-slice clinical CT scanner at four different dose levels [CTDIvol =4, 8, 12, 16 (mGy)]. Both filtered backprojection (FBP) and MBIR (Veo(®), GE Healthcare, Waukesha, WI) were used for image reconstruction and results were compared with one another. Eight test objects in the phantom with contrast levels ranging from 13 to 1710 HU were used to assess spatial resolution. The axial spatial resolution was quantified with the point spread function (PSF), while the z resolution was quantified with the slice sensitivity profile. Both were measured locally on the test objects and in the image domain. The dependence of spatial resolution on contrast and dose levels was studied. The study also features a systematic investigation of the potential trade-off between spatial resolution and locally defined noise and their joint impact on the overall image quality, which was quantified by the image domain-based channelized Hotelling observer (CHO) detectability index d'.

RESULTS

(1) The axial spatial resolution of MBIR depends on both radiation dose level and image contrast level, whereas it is supposedly independent of these two factors in FBP. The axial spatial resolution of MBIR always improved with an increasing radiation dose level and/or contrast level. (2) The axial spatial resolution of MBIR became equivalent to that of FBP at some transitional contrast level, above which MBIR demonstrated superior spatial resolution than FBP (and vice versa); the value of this transitional contrast highly depended on the dose level. (3) The PSFs of MBIR could be approximated as Gaussian functions with reasonably good accuracy. (4) Thez resolution of MBIR showed similar contrast and dose dependence. (5) Noise standard deviation assessed on the edges of objects demonstrated a trade-off with spatial resolution in MBIR. (5) When both spatial resolution and image noise were considered using the CHO analysis, MBIR led to significant improvement in the overall CT image quality for both high and low contrast detection tasks at both standard and low dose levels.

CONCLUSIONS

Due to the intrinsic nonlinearity of the MBIR method, many well-known CT spatial resolution and noise properties have been modified. In particular, dose dependence and contrast dependence have been introduced to the spatial resolution of CT images by MBIR. The method has also introduced some novel noise-resolution trade-off not seen in traditional CT images. While the benefits of MBIR regarding the overall image quality, as demonstrated in this work, are significant, the optimal use of this method in clinical practice demands a thorough understanding of its unique physical characteristics.

Tensor-based formulation and nuclear norm regularization for multienergy computed tomography

The development of energy selective, photon counting X-ray detectors allows for a wide range of new possibilities in the area of computed tomographic image formation. Under the assumption of perfect energy resolution, here we propose a tensor-based iterative algorithm that simultaneously reconstructs the X-ray attenuation distribution for each energy. We use a multilinear image model rather than a more standard stacked vector representation in order to develop novel tensor-based regularizers. In particular, we model the multispectral unknown as a three-way tensor where the first two dimensions are space and the third dimension is energy. This approach allows for the design of tensor nuclear norm regularizers, which like its 2D counterpart, is a convex function of the multispectral unknown. The solution to the resulting convex optimization problem is obtained using an alternating direction method of multipliers approach. Simulation results show that the generalized tensor nuclear norm can be used as a standalone regularization technique for the energy selective (spectral) computed tomography problem and when combined with total variation regularization it enhances the regularization capabilities especially at low energy images where the effects of noise are most prominent.

Regularization design in penalized maximum-likelihood image reconstruction for lesion detection in 3D PET

Detecting cancerous lesions is a major clinical application in emission tomography. In previous work, we have studied penalized maximum-likelihood (PML) image reconstruction for the detection task and proposed a method to design a shift-invariant quadratic penalty function to maximize detectability of a lesion at a known location in a two dimensional image. Here we extend the regularization design to maximize detectability of lesions at unknown locations in fully 3D PET. We used a multiview channelized Hotelling observer (mvCHO) to assess the lesion detectability in 3D images to mimic the condition where a human observer examines three orthogonal views of a 3D image for lesion detection. We derived simplified theoretical expressions that allow fast prediction of the detectability of a 3D lesion. The theoretical results were used to design the regularization in PML reconstruction to improve lesion detectability. We conducted computer-based Monte Carlo simulations to compare the optimized penalty with the conventional penalty for detecting lesions of various sizes. Only true coincidence events were simulated. Lesion detectability was also assessed by two human observers, whose performances agree well with that of the mvCHO. Both the numerical observer and human observer results showed a statistically significant improvement in lesion detection by using the proposed penalty function compared to using the conventional penalty function.

Model-Based Iterative Reconstruction for Dual-Energy X-Ray CT Using a Joint Quadratic Likelihood Model

Dual-energy X-ray CT (DECT) has the potential to improve contrast and reduce artifacts as compared to traditional CT. Moreover, by applying model-based iterative reconstruction (MBIR) to dual-energy data, one might also expect to reduce noise and improve resolution. However, the direct implementation of dual-energy MBIR requires the use of a nonlinear forward model, which increases both complexity and computation. Alternatively, simplified forward models have been used which treat the material-decomposed channels separately, but these approaches do not fully account for the statistical dependencies in the channels. In this paper, we present a method for joint dual-energy MBIR (JDE-MBIR), which simplifies the forward model while still accounting for the complete statistical dependency in the material-decomposed sinogram components. The JDE-MBIR approach works by using a quadratic approximation to the polychromatic log-likelihood and a simple but exact nonnegativity constraint in the image domain. We demonstrate that our method is particularly effective when the DECT system uses fast kVp switching, since in this case the model accounts for the inaccuracy of interpolated sinogram entries. Both phantom and clinical results show that the proposed model produces images that compare favorably in quality to previous decomposition-based methods, including FBP and other statistical iterative approaches.

Iterative image reconstruction for sparse-view CT using normal-dose image induced total variation prior

X-ray computed tomography (CT) iterative image reconstruction from sparse-view projection data has been an important research topic for radiation reduction in clinic. In this paper, to relieve the requirement of misalignment reduction operation of the prior image constrained compressed sensing (PICCS) approach introduced by Chen et al, we present an iterative image reconstruction approach for sparse-view CT using a normal-dose image induced total variation (ndiTV) prior. The associative objective function of the present approach is constructed under the penalized weighed least-square (PWLS) criteria, which contains two terms, i.e., the weighted least-square (WLS) fidelity and the ndiTV prior, and is referred to as "PWLS-ndiTV". Specifically, the WLS fidelity term is built based on an accurate relationship between the variance and mean of projection data in the presence of electronic background noise. The ndiTV prior term is designed to reduce the influence of the misalignment between the desired- and prior- image by using a normal-dose image induced non-local means (ndiNLM) filter. Subsequently, a modified steepest descent algorithm is adopted to minimize the associative objective function. Experimental results on two different digital phantoms and an anthropomorphic torso phantom show that the present PWLS-ndiTV approach for sparse-view CT image reconstruction can achieve noticeable gains over the existing similar approaches in terms of noise reduction, resolution-noise tradeoff, and low-contrast object detection.

A model based iterative reconstruction algorithm for high angle annular dark field-scanning transmission electron microscope (HAADF-STEM) tomography

High angle annular dark field (HAADF)-scanning transmission electron microscope (STEM) data is increasingly being used in the physical sciences to research materials in 3D because it reduces the effects of Bragg diffraction seen in bright field TEM data. Typically, tomographic reconstructions are performed by directly applying either filtered back projection (FBP) or the simultaneous iterative reconstruction technique (SIRT) to the data. Since HAADF-STEM tomography is a limited angle tomography modality with low signal to noise ratio, these methods can result in significant artifacts in the reconstructed volume. In this paper, we develop a model based iterative reconstruction algorithm for HAADF-STEM tomography. We combine a model for image formation in HAADF-STEM tomography along with a prior model to formulate the tomographic reconstruction as a maximum a posteriori probability (MAP) estimation problem. Our formulation also accounts for certain missing measurements by treating them as nuisance parameters in the MAP estimation framework. We adapt the iterative coordinate descent algorithm to develop an efficient method to minimize the corresponding MAP cost function. Reconstructions of simulated as well as experimental data sets show results that are superior to FBP and SIRT reconstructions, significantly suppressing artifacts and enhancing contrast.

Quantitative statistical methods for image quality assessment

Quantitative measures of image quality and reliability are critical for both qualitative interpretation and quantitative analysis of medical images. While, in theory, it is possible to analyze reconstructed images by means of Monte Carlo simulations using a large number of noise realizations, the associated computational burden makes this approach impractical. Additionally, this approach is less meaningful in clinical scenarios, where multiple noise realizations are generally unavailable. The practical alternative is to compute closed-form analytical expressions for image quality measures. The objective of this paper is to review statistical analysis techniques that enable us to compute two key metrics: resolution (determined from the local impulse response) and covariance. The underlying methods include fixed-point approaches, which compute these metrics at a fixed point (the unique and stable solution) independent of the iterative algorithm employed, and iteration-based approaches, which yield results that are dependent on the algorithm, initialization, and number of iterations. We also explore extensions of some of these methods to a range of special contexts, including dynamic and motion-compensated image reconstruction. While most of the discussed techniques were developed for emission tomography, the general methods are extensible to other imaging modalities as well. In addition to enabling image characterization, these analysis techniques allow us to control and enhance imaging system performance. We review practical applications where performance improvement is achieved by applying these ideas to the contexts of both hardware (optimizing scanner design) and image reconstruction (designing regularization functions that produce uniform resolution or maximize task-specific figures of merit).

Direct Reconstruction of CT-based Attenuation Correction Images for PET with Cluster-Based Penalties

Extremely low-dose CT acquisitions for the purpose of PET attenuation correction will have a high level of noise and biasing artifacts due to factors such as photon starvation. This work explores a priori knowledge appropriate for CT iterative image reconstruction for PET attenuation correction. We investigate the maximum a posteriori (MAP) framework with cluster-based, multinomial priors for the direct reconstruction of the PET attenuation map. The objective function for direct iterative attenuation map reconstruction was modeled as a Poisson log-likelihood with prior terms consisting of quadratic (Q) and mixture (M) distributions. The attenuation map is assumed to have values in 4 clusters: air+background, lung, soft tissue, and bone. Under this assumption, the MP was a mixture probability density function consisting of one exponential and three Gaussian distributions. The relative proportion of each cluster was jointly estimated during each voxel update of direct iterative coordinate decent (dICD) method. Noise-free data were generated from NCAT phantom and Poisson noise was added. Reconstruction with FBP (ramp filter) was performed on the noise-free (ground truth) and noisy data. For the noisy data, dICD reconstruction was performed with the combination of different prior strength parameters (β and γ) of Q- and M-penalties. The combined quadratic and mixture penalties reduces the RMSE by 18.7% compared to post-smoothed iterative reconstruction and only 0.7% compared to quadratic alone. For direct PET attenuation map reconstruction from ultra-low dose CT acquisitions, the combination of quadratic and mixture priors offers regularization of both variance and bias and is a potential method to derive attenuation maps with negligible patient dose. However, the small improvement in quantitative accuracy relative to the substantial increase in algorithm complexity does not currently justify the use of mixture-based PET attenuation priors for reconstruction of CT images for PET attenuation correction.

Characterization of statistical prior image constrained compressed sensing (PICCS): II. Application to dose reduction

PURPOSE

The ionizing radiation imparted to patients during computed tomography exams is raising concerns. This paper studies the performance of a scheme called dose reduction using prior image constrained compressed sensing (DR-PICCS). The purpose of this study is to characterize the effects of a statistical model of x-ray detection in the DR-PICCS framework and its impact on spatial resolution.

METHODS

Both numerical simulations with known ground truth and in vivo animal dataset were used in this study. In numerical simulations, a phantom was simulated with Poisson noise and with varying levels of eccentricity. Both the conventional filtered backprojection (FBP) and the PICCS algorithms were used to reconstruct images. In PICCS reconstructions, the prior image was generated using two different denoising methods: a simple Gaussian blur and a more advanced diffusion filter. Due to the lack of shift-invariance in nonlinear image reconstruction such as the one studied in this paper, the concept of local spatial resolution was used to study the sharpness of a reconstructed image. Specifically, a directional metric of image sharpness, the so-called pseudopoint spread function (pseudo-PSF), was employed to investigate local spatial resolution.

RESULTS

In the numerical studies, the pseudo-PSF was reduced from twice the voxel width in the prior image down to less than 1.1 times the voxel width in DR-PICCS reconstructions when the statistical model was not included. At the same noise level, when statistical weighting was used, the pseudo-PSF width in DR-PICCS reconstructed images varied between 1.5 and 0.75 times the voxel width depending on the direction along which it was measured. However, this anisotropy was largely eliminated when the prior image was generated using diffusion filtering; the pseudo-PSF width was reduced to below one voxel width in that case. In the in vivo study, a fourfold improvement in CNR was achieved while qualitatively maintaining sharpness; images also had a qualitatively more uniform noise spatial distribution when including a statistical model.

CONCLUSIONS

DR-PICCS enables to reconstruct CT images with lower noise than FBP and the loss of spatial resolution can be mitigated to a large extent. The introduction of statistical modeling in DR-PICCS may improve some noise characteristics, but it also leads to anisotropic spatial resolution properties. A denoising method, such as the directional diffusion filtering, has been demonstrated to reduce anisotropy in spatial resolution effectively when it was combined with DR-PICCS with statistical modeling.

Information-theoretic discrepancy based iterative reconstructions (IDIR) for polychromatic x-ray tomography

PURPOSE

X-ray photons generated from a typical x-ray source for clinical applications exhibit a broad range of wavelengths, and the interactions between individual particles and biological substances depend on particles' energy levels. Most existing reconstruction methods for transmission tomography, however, neglect this polychromatic nature of measurements and rely on the monochromatic approximation. In this study, we developed a new family of iterative methods that incorporates the exact polychromatic model into tomographic image recovery, which improves the accuracy and quality of reconstruction.

METHODS

The generalized information-theoretic discrepancy (GID) was employed as a new metric for quantifying the distance between the measured and synthetic data. By using special features of the GID, the objective function for polychromatic reconstruction which contains a double integral over the wavelength and the trajectory of incident x-rays was simplified to a paraboloidal form without using the monochromatic approximation. More specifically, the original GID was replaced with a surrogate function with two auxiliary, energy-dependent variables. Subsequently, the alternating minimization technique was applied to solve the double minimization problem. Based on the optimization transfer principle, the objective function was further simplified to the paraboloidal equation, which leads to a closed-form update formula. Numerical experiments on the beam-hardening correction and material-selective reconstruction were conducted to compare and assess the performance of conventional methods and the proposed algorithms.

RESULTS

The authors found that the GID determines the distance between its two arguments in a flexible manner. In this study, three groups of GIDs with distinct data representations were considered. The authors demonstrated that one type of GIDs that comprises "raw" data can be viewed as an extension of existing statistical reconstructions; under a particular condition, the GID is equivalent to the Poisson log-likelihood function. The newly proposed GIDs of the other two categories consist of log-transformed measurements, which have the advantage of imposing linearized penalties over multiple discrepancies. For all proposed variants of the GID, the aforementioned strategy was used to obtain a closed-form update equation. Even though it is based on the exact polychromatic model, the derived algorithm bears a structural resemblance to conventional methods based on the monochromatic approximation. The authors named the proposed approach as information-theoretic discrepancy based iterative reconstructions (IDIR). In numerical experiments, IDIR with raw data converged faster than previously known statistical reconstruction methods. IDIR with log-transformed data exhibited superior reconstruction quality and faster convergence speed compared with conventional methods and their variants.

CONCLUSIONS

The authors' new framework for tomographic reconstruction allows iterative inversion of the polychromatic data model. The primary departure from the traditional iterative reconstruction was the employment of the GID as a new metric for quantifying the inconsistency between the measured and synthetic data. The proposed methods outperformed not only conventional methods based on the monochromatic approximation but also those based on the polychromatic model. The authors have observed that the GID is a very flexible means to design an objective function for iterative reconstructions. Hence, the authors expect that the proposed IDIR framework will also be applicable to other challenging tasks.

Anomaly Detection and Artifact Recovery in PET Attenuation-Correction Images Using the Likelihood Function

In dual modality PET/CT, CT data are used to generate the attenuation correction applied in the reconstruction of the PET emission image. This requires converting the CT image into a 511-keV attenuation map. Algorithms for making this transformation require assumptions about the makeup of material within the patient. Anomalous material such as contrast agent administered to enhance the CT scan confounds conversion algorithms and has been observed to result in inaccuracies, i.e., inconsistencies with the true 511-keV attenuation present at the time of the PET emission scan. These attenuation artifacts carry through to the final attenuation-corrected PET emission image and can resemble diseased tissue. We propose an approach to correcting this problem that employs the attenuation information carried by the PET emission data. A likelihood-based algorithm for identifying and correcting of contrast is presented and tested. The algorithm exploits the fact that contrast artifacts manifest as too-high attenuation values in an otherwise high quality attenuation image. In a separate study, the performance of the loglikelihood as an objective-function component of a detection/correction algorithm, independent of any particular algorithm was mapped out for several imaging scenarios as a function of statistical noise. Both the full algorithm and the loglikelihood performed well in studies with simulated data. Additional studies including those with patient data are required to fully understand their capabilities.

Gaussian process inference for estimating pharmacokinetic parameters of dynamic contrast-enhanced MR images

In this paper, we propose a new pharmacokinetic model for parameter estimation of dynamic contrast-enhanced (DCE) MRI by using Gaussian process inference. Our model is based on the Tofts dual-compartment model for the description of tracer kinetics and the observed time series from DCE-MRI is treated as a Gaussian stochastic process. The parameter estimation is done through a maximum likelihood approach and we propose a variant of the coordinate descent method to solve this likelihood maximization problem. The new model was shown to outperform a baseline method on simulated data. Parametric maps generated on prostate DCE data with the new model also provided better enhancement of tumors, lower intensity on false positives, and better boundary delineation when compared with the baseline method. New statistical parameter maps from the process model were also found to be informative, particularly when paired with the PK parameter maps.

Characterization of statistical prior image constrained compressed sensing. I. Applications to time-resolved contrast-enhanced CT

PURPOSE

Prior image constrained compressed sensing (PICCS) is an image reconstruction framework that takes advantage of a prior image to improve the image quality of CT reconstructions. An interesting question that remains to be investigated is whether or not the introduction of a statistical model of the photon detection in the PICCS reconstruction framework can improve the performance of the algorithm when dealing with high noise projection datasets. The goal of the research presented in this paper is to characterize the noise properties of images reconstructed using PICCS with and without statistical modeling. This paper investigates these properties in the clinical context of time-resolved contrast-enhanced CT.

METHODS

Both numerical phantom studies and an Institutional Review Board approved human subject study were used in this research. The conventional filtered backprojection (FBP), and PICCS with and without the statistical model were applied to each dataset. The prior image used in PICCS was generated by averaging over FBP reconstructions from different time frames of the time-resolved CT exam, thus reducing the noise level. Numerical studies were used to evaluate if the noise characteristics are altered for varying levels of noise, as well as for different object shapes. The dataset acquired in vivo was used to verify that the conclusions reached from numerical studies translate adequately to a clinical case. The results were analyzed using a variety of qualitative and quantitative metrics such as the universal image quality index, spatial maps of the noise standard deviations, the noise uniformity, the noise power spectrum, and the model-observer detectability.

RESULTS

The noise characteristics of PICCS were shown to depend on the noise level contained in the data, the level of eccentricity of the object, and whether or not the statistical model was applied. Most differences in the characteristics were observed in the regime of low incident x-ray fluence. No substantial difference was observed between PICCS with and without statistics in the high fluence domain. Objects with a semi-major axis ratio below 0.85 were more accurately reconstructed with lower noise using the statistical implementation. Above that range, for mostly circular objects, the PICCS implementation without the statistical model yielded more accurate images and a lower noise level. At all levels of eccentricity, the noise spatial distribution was more uniform and the model-observer detectability was greater for PICCS with the statistical model. The human subject study was consistent with the results obtained using numerical simulations.

CONCLUSIONS

For mildly eccentric objects in the low noise regime, PICCS without the noise model yielded equal or better noise level and image quality than the statistical formulation. However, in a vast majority of cases, images reconstructed using statistical PICCS have a noise power spectrum that facilitated the detection of model lesions. The inclusion of a statistical model in the PICCS framework does not always result in improved noise characteristics.

Noise models for low counting rate coherent diffraction imaging

Coherent diffraction imaging (CDI) is a lens-less microscopy method that extracts the complex-valued exit field from intensity measurements alone. It is of particular importance for microscopy imaging with diffraction set-ups where high quality lenses are not available. The inversion scheme allowing the phase retrieval is based on the use of an iterative algorithm. In this work, we address the question of the choice of the iterative process in the case of data corrupted by photon or electron shot noise. Several noise models are presented and further used within two inversion strategies, the ordered subset and the scaled gradient. Based on analytical and numerical analysis together with Monte-Carlo studies, we show that any physical interpretations drawn from a CDI iterative technique require a detailed understanding of the relationship between the noise model and the used inversion method. We observe that iterative algorithms often assume implicitly a noise model. For low counting rates, each noise model behaves differently. Moreover, the used optimization strategy introduces its own artefacts. Based on this analysis, we develop a hybrid strategy which works efficiently in the absence of an informed initial guess. Our work emphasises issues which should be considered carefully when inverting experimental data.

Noise spatial nonuniformity and the impact of statistical image reconstruction in CT myocardial perfusion imaging

PURPOSE

To achieve high temporal resolution in CT myocardial perfusion imaging (MPI), images are often reconstructed using filtered backprojection (FBP) algorithms from data acquired within a short-scan angular range. However, the variation in the central angle from one time frame to the next in gated short scans has been shown to create detrimental partial scan artifacts when performing quantitative MPI measurements. This study has two main purposes. (1) To demonstrate the existence of a distinct detrimental effect in short-scan FBP, i.e., the introduction of a nonuniform spatial image noise distribution; this nonuniformity can lead to unexpectedly high image noise and streaking artifacts, which may affect CT MPI quantification. (2) To demonstrate that statistical image reconstruction (SIR) algorithms can be a potential solution to address the nonuniform spatial noise distribution problem and can also lead to radiation dose reduction in the context of CT MPI.

METHODS

Projection datasets from a numerically simulated perfusion phantom and an in vivo animal myocardial perfusion CT scan were used in this study. In the numerical phantom, multiple realizations of Poisson noise were added to projection data at each time frame to investigate the spatial distribution of noise. Images from all datasets were reconstructed using both FBP and SIR reconstruction algorithms. To quantify the spatial distribution of noise, the mean and standard deviation were measured in several regions of interest (ROIs) and analyzed across time frames. In the in vivo study, two low-dose scans at tube currents of 25 and 50 mA were reconstructed using FBP and SIR. Quantitative perfusion metrics, namely, the normalized upslope (NUS), myocardial blood volume (MBV), and first moment transit time (FMT), were measured for two ROIs and compared to reference values obtained from a high-dose scan performed at 500 mA.

RESULTS

Images reconstructed using FBP showed a highly nonuniform spatial distribution of noise. This spatial nonuniformity led to large fluctuations in the temporal direction. In the numerical phantom study, the level of noise was shown to vary by as much as 87% within a given image, and as much as 110% between different time frames for a ROI far from isocenter. The spatially nonuniform noise pattern was shown to correlate with the source trajectory and the object structure. In contrast, images reconstructed using SIR showed a highly uniform spatial distribution of noise, leading to smaller unexpected noise fluctuations in the temporal direction when a short scan angular range was used. In the numerical phantom study, the noise varied by less than 37% within a given image, and by less than 20% between different time frames. Also, the noise standard deviation in SIR images was on average half of that of FBP images. In the in vivo studies, the deviation observed between quantitative perfusion metrics measured from low-dose scans and high-dose scans was mitigated when SIR was used instead of FBP to reconstruct images.

CONCLUSIONS

(1) Images reconstructed using FBP suffered from nonuniform spatial noise levels. This nonuniformity is another manifestation of the detrimental effects caused by short-scan reconstruction in CT MPI. (2) Images reconstructed using SIR had a much lower and more uniform noise level and thus can be used as a potential solution to address the FBP nonuniformity. (3) Given the improvement in the accuracy of the perfusion metrics when using SIR, it may be desirable to use a statistical reconstruction framework to perform low-dose dynamic CT MPI.

A parametric level-set approach to simultaneous object identification and background reconstruction for dual-energy computed tomography

Dual-energy computerized tomography has gained great interest because of its ability to characterize the chemical composition of a material rather than simply providing relative attenuation images as in conventional tomography. The purpose of this paper is to introduce a novel polychromatic dual-energy processing algorithm, with an emphasis on detection and characterization of piecewise constant objects embedded in an unknown cluttered background. Physical properties of the objects, particularly the Compton scattering and photoelectric absorption coefficients, are assumed to be known with some level of uncertainty. Our approach is based on a level-set representation of the characteristic function of the object and encompasses a number of regularization techniques for addressing both the prior information we have concerning the physical properties of the object and the fundamental physics-based limitations associated with our ability to jointly recover the Compton scattering and photoelectric absorption properties of the scene. In the absence of an object with appropriate physical properties, our approach returns a null characteristic function and, thus, can be viewed as simultaneously solving the detection and characterization problems. Unlike the vast majority of methods that define the level-set function nonparametrically, i.e., as a dense set of pixel values, we define our level set parametrically via radial basis functions and employ a Gauss-Newton-type algorithm for cost minimization. Numerical results show that the algorithm successfully detects objects of interest, finds their shape and location, and gives an adequate reconstruction of the background.

Iterative reconstruction methods in X-ray CT

Iterative reconstruction (IR) methods have recently re-emerged in transmission x-ray computed tomography (CT). They were successfully used in the early years of CT, but given up when the amount of measured data increased because of the higher computational demands of IR compared to analytical methods. The availability of large computational capacities in normal workstations and the ongoing efforts towards lower doses in CT have changed the situation; IR has become a hot topic for all major vendors of clinical CT systems in the past 5 years. This review strives to provide information on IR methods and aims at interested physicists and physicians already active in the field of CT. We give an overview on the terminology used and an introduction to the most important algorithmic concepts including references for further reading. As a practical example, details on a model-based iterative reconstruction algorithm implemented on a modern graphics adapter (GPU) are presented, followed by application examples for several dedicated CT scanners in order to demonstrate the performance and potential of iterative reconstruction methods. Finally, some general thoughts regarding the advantages and disadvantages of IR methods as well as open points for research in this field are discussed.

Using spatial prior knowledge in the spectral fitting of MRS images

We propose a Bayesian smoothness prior in the spectral fitting of MRS images which can be used in addition to commonly employed prior knowledge. By combining a frequency-domain model for the free induction decay with a Gaussian Markov random field prior, a new optimization objective is derived that encourages smooth parameter maps. Using a particular parameterization of the prior, smooth damping, frequency and phase maps can be obtained whilst preserving sharp spatial features in the amplitude map. A Monte Carlo study based on two sets of simulated data demonstrates that the variance of the estimated parameter maps can be reduced considerably, even below the Cramér-Rao lower bound, when using spatial prior knowledge. Long-TE (1)H MRSI at 1.5 T of a patient with a brain tumor shows that the use of the spatial prior resolves the overlapping peaks of choline and creatine when a single voxel method fails to do so. Improved and detailed metabolic maps can be derived from high-spatial-resolution, short-TE (1)H MRSI at 3 T. Finally, the evaluation of four series of long-TE brain MRSI data with various signal-to-noise ratios shows the general benefit of the proposed approach.

Effects of the penalty on the penalized weighted least-squares image reconstruction for low-dose CBCT

Statistical iterative reconstruction (SIR) algorithms have shown potential to substantially improve low-dose cone-beam CT (CBCT) image quality. The penalty term plays an important role in determining the performance of SIR algorithms. In this work, we quantitatively evaluate the impact of the penalties on the performance of a statistics-based penalized weighted least-squares (PWLS) iterative reconstruction algorithm for improving the image quality of low-dose CBCT. Three different edge-preserving penalty terms, exponential form anisotropic quadratic (AQ) penalty (PWLS-Exp), inverse square form AQ penalty (PWLS-InverseSqr) and total variation penalty (PWLS-TV), were compared against the conventional isotropic quadratic form penalty (PWLS-Iso) using both computer simulation and experimental studies. Noise in low-dose CBCT can be substantially suppressed by the PWLS reconstruction algorithm and edges are well preserved by both AQ- and TV-based penalty terms. The noise-resolution tradeoff measurement shows that the PWLS-Exp exhibits the best spatial resolution of all the three anisotropic penalty terms at matched noise level for reconstructing high-contrast objects. For the reconstruction of low-contrast objects, the TV-based penalty outperforms the AQ-based one with better resolution preservation at matched noise levels. Different penalty terms may be used for better edge preservation at different targeted contrast levels.

Penalized maximum likelihood reconstruction for improved microcalcification detection in breast tomosynthesis

We examined the application of an iterative penalized maximum likelihood (PML) reconstruction method for improved detectability of microcalcifications (MCs) in digital breast tomosynthesis (DBT). Localized receiver operating characteristic (LROC) psychophysical studies with human observers and 2-D image slices were conducted to evaluate the performance of this reconstruction method and to compare its performance against the commonly used Feldkamp FBP algorithm. DBT projections were generated using rigorous computer simulations that included accurate modeling of the noise and detector blur. Acquisition dose levels of 0.7, 1.0, and 1.5 mGy in a 5-cm-thick compressed breast were tested. The defined task was to localize and detect MC clusters consisting of seven MCs. The individual MC diameter was 150 μm. Compressed-breast phantoms derived from CT images of actual mastectomy specimens provided realistic background structures for the detection task. Four observers each read 98 test images for each combination of reconstruction method and acquisition dose. All observers performed better with the PML images than with the FBP images. With the acquisition dose of 0.7 mGy, the average areas under the LROC curve (A(L)) for the PML and FBP algorithms were 0.69 and 0.43, respectively. For the 1.0-mGy dose, the values of A(L) were 0.93 (PML) and 0.7 (FBP), while the 1.5-mGy dose resulted in areas of 1.0 and 0.9, respectively, for the PML and FBP algorithms. A 2-D analysis of variance applied to the individual observer areas showed statistically significant differences (at a significance level of 0.05) between the reconstruction strategies at all three dose levels. There were no significant differences in observer performance for any of the dose levels.

Iterative reconstruction in X-ray fluorescence tomography based on radon inversion

We describe a new approach for the inversion of the generalized attenuated radon transform in X-ray fluorescence computed tomography (XFCT). The approach consists of using the radon inverse as an approximation for the actual one, followed by an iterative refinement. Also, we analyze the problem of retrieving the attenuation map directly from the emission data, giving rise to a novel alternating method for the solution. We applied our approach to real and simulated XFCT data and compared its performance to previous inversion algorithms for the problem, showing its main advantages: better images than those obtained by other analytic methods and much faster than iterative methods in the discrete setting.

Tomographic reconstruction of gated data acquisition using DFT basis functions

In image reconstruction gated acquisition is often used in order to deal with blur caused by organ motion in the resulting images. However, this is achieved almost inevitably at the expense of reduced signal-to-noise ratio in the acquired data. In this work, we propose a reconstruction procedure for gated images based upon use of discrete Fourier transform (DFT) basis functions, wherein the temporal activity at each spatial location is regulated by a Fourier representation. The gated images are then reconstructed through determination of the coefficients of the Fourier representation. We demonstrate this approach in the context of single photon emission computed tomography (SPECT) for cardiac imaging, which is often hampered by the increased noise due to gating and other degrading factors. We explore two different reconstruction algorithms, one is a penalized least-square approach and the other is a maximum a posteriori approach. In our experiments, we conducted a quantitative evaluation of the proposed approach using Monte Carlo simulated SPECT imaging. The results demonstrate that use of DFT-basis functions in gated imaging can improve the accuracy of the reconstruction. As a preliminary demonstration, we also tested this approach on a set of clinical acquisition.

Fast model-based X-ray CT reconstruction using spatially nonhomogeneous ICD optimization

Recent applications of model-based iterative reconstruction (MBIR) algorithms to multislice helical CT reconstructions have shown that MBIR can greatly improve image quality by increasing resolution as well as reducing noise and some artifacts. However, high computational cost and long reconstruction times remain as a barrier to the use of MBIR in practical applications. Among the various iterative methods that have been studied for MBIR, iterative coordinate descent (ICD) has been found to have relatively low overall computational requirements due to its fast convergence. This paper presents a fast model-based iterative reconstruction algorithm using spatially nonhomogeneous ICD (NH-ICD) optimization. The NH-ICD algorithm speeds up convergence by focusing computation where it is most needed. The NH-ICD algorithm has a mechanism that adaptively selects voxels for update. First, a voxel selection criterion VSC determines the voxels in greatest need of update. Then a voxel selection algorithm VSA selects the order of successive voxel updates based upon the need for repeated updates of some locations, while retaining characteristics for global convergence. In order to speed up each voxel update, we also propose a fast 1-D optimization algorithm that uses a quadratic substitute function to upper bound the local 1-D objective function, so that a closed form solution can be obtained rather than using a computationally expensive line search algorithm. We examine the performance of the proposed algorithm using several clinical data sets of various anatomy. The experimental results show that the proposed method accelerates the reconstructions by roughly a factor of three on average for typical 3-D multislice geometries.

Performance comparison between total variation (TV)-based compressed sensing and statistical iterative reconstruction algorithms

Of all available reconstruction methods, statistical iterative reconstruction algorithms appear particularly promising since they enable accurate physical noise modeling. The newly developed compressive sampling/compressed sensing (CS) algorithm has shown the potential to accurately reconstruct images from highly undersampled data. The CS algorithm can be implemented in the statistical reconstruction framework as well. In this study, we compared the performance of two standard statistical reconstruction algorithms (penalized weighted least squares and q-GGMRF) to the CS algorithm. In assessing the image quality using these iterative reconstructions, it is critical to utilize realistic background anatomy as the reconstruction results are object dependent. A cadaver head was scanned on a Varian Trilogy system at different dose levels. Several figures of merit including the relative root mean square error and a quality factor which accounts for the noise performance and the spatial resolution were introduced to objectively evaluate reconstruction performance. A comparison is presented between the three algorithms for a constant undersampling factor comparing different algorithms at several dose levels. To facilitate this comparison, the original CS method was formulated in the framework of the statistical image reconstruction algorithms. Important conclusions of the measurements from our studies are that (1) for realistic neuro-anatomy, over 100 projections are required to avoid streak artifacts in the reconstructed images even with CS reconstruction, (2) regardless of the algorithm employed, it is beneficial to distribute the total dose to more views as long as each view remains quantum noise limited and (3) the total variation-based CS method is not appropriate for very low dose levels because while it can mitigate streaking artifacts, the images exhibit patchy behavior, which is potentially harmful for medical diagnosis.

FDG-PET parametric imaging by total variation minimization

Parametric imaging of the cerebral metabolic rate for glucose (CMRGlc) using [(18)F]-fluoro deoxyglucose positron emission tomography is considered. Traditional imaging is hindered due to low signal-to-noise ratios at individual voxels. We propose to minimize the total variation of the tracer uptake rates while requiring good fit of traditional Patlak equations. This minimization guarantees spatial homogeneity within brain regions and good distinction between brain regions. Brain phantom simulations demonstrate significant improvement in quality of images by the proposed method as compared to Patlak images with post-filtering using Gaussian or median filters.

Clustering dynamic PET images on the Gaussian distributed sinogram domain

Segmentation of dynamic PET images is an important preprocessing step for kinetic parameter estimation. The time activity curve (TAC) of individual pixels has very low signal-to-noise ratio (SNR). Therefore, the kinetic parameters estimated from the TAC of an individual pixel may not be accurate, and these estimations may have very high spatial variance. To alleviate this problem, pixels with similar kinetic characteristics are clustered into regions, and TACs of pixels within each region are averaged to increase SNR. It has recently been shown that clustering dynamic PET images in the sinogram domain is better than clustering them in the reconstructed image domain [M.E. Kamasak, B. Bayraktar, Clustering dynamic PET images on the projection domain, IEEE Trans. Nucl. Sci. 54 (3) (June 2007) 496-503.]. In that study, the sinograms are assumed to have Poisson distribution. The clusters and TACs of the clusters are then chosen to maximize the posterior probability of the measured sinograms. Although the raw sinogram data are Poisson distributed, the sinogram data corrected for scatter, randoms, attenuation etc. are not Poisson distributed anymore. In this paper, we describe how to cluster dynamic PET images on the sinogram domain when the sinograms are Gaussian distributed.

A survey of approaches for direct parametric image reconstruction in emission tomography

The quantitative data obtained by emission tomography are decoded using a number of techniques and methods in sequence to provide physiological information. Conventionally, the data are reconstructed to produce a series of static images. Then, pharmacokinetic modeling techniques are applied, and kinetic parameters that have physiological or functional significance are derived. Although it is possible to optimize each estimation step in this process, many simplifying assumptions have to be introduced to make the methods that are used practicable. Published research has shown that if the kinetic parameters are estimated directly from the measured data, the parametric images will have higher quality and lower mean-squared error than if this was done indirectly. This review highlights some aspects of the methods that have been proposed for such direct estimation of pharmacokinetic information from raw emission data.

Noniterative sequential weighted least squares algorithm for positron emission tomography reconstruction

This paper proposes a new sequential weighted least squares (SWLS) method for positron emission tomography (PET) reconstruction. The SWLS algorithm is noniterative and can be considered as equivalent to the penalized WLS (PWLS) method under certain initial conditions. However, a full implementation of SWLS is computationally intensive. To overcome this problem, we propose a simplified SWLS as a reasonable alternative to the SWLS. The performance of this SWLS method is evaluated in experiments using both simulated and clinical data. The results show that the method can be advantageously compared with the original SWLS both in computation time and reconstruction quality.

Joint model of motion and anatomy for PET image reconstruction

Anatomy-based positron emission tomography (PET) image enhancement techniques have been shown to have the potential for improving PET image quality. However, these techniques assume an accurate alignment between the anatomical and the functional images, which is not always valid when imaging the chest due to respiratory motion. In this article, we present a joint model of both motion and anatomical information by integrating a motion-incorporated PET imaging system model with an anatomy-based maximum a posteriori image reconstruction algorithm. The mismatched anatomical information due to motion can thus be effectively utilized through this joint model. A computer simulation and a phantom study were conducted to assess the efficacy of the joint model, whereby motion and anatomical information were either modeled separately or combined. The reconstructed images in each case were compared to corresponding reference images obtained using a quadratic image prior based maximum a posteriori reconstruction algorithm for quantitative accuracy. Results of these studies indicated that while modeling anatomical information or motion alone improved the PET image quantitation accuracy, a larger improvement in accuracy was achieved when using the joint model. In the computer simulation study and using similar image noise levels, the improvement in quantitation accuracy compared to the reference images was 5.3% and 19.8% when using anatomical or motion information alone, respectively, and 35.5% when using the joint model. In the phantom study, these results were 5.6%, 5.8%, and 19.8%, respectively. These results suggest that motion compensation is important in order to effectively utilize anatomical information in chest imaging using PET. The joint motion-anatomy model presented in this paper provides a promising solution to this problem.

Block-iterative Fisher scoring algorithms for maximum penalized likelihood image reconstruction in emission tomography

This paper introduces and evaluates a block-iterative Fisher scoring (BFS) algorithm. The algorithm provides regularized estimation in tomographic models of projection data with Poisson variability. Regularization is achieved by penalized likelihood with a general quadratic penalty. Local convergence of the block-iterative algorithm is proven under conditions that do not require iteration dependent relaxation. We show that, when the algorithm converges, it converges to the unconstrained maximum penalized likelihood (MPL) solution. Simulation studies demonstrate that, with suitable choice of relaxation parameter and restriction of the algorithm to respect nonnegative constraints, the BFS algorithm provides convergence to the constrained MPL solution. Constrained BFS often attains a maximum penalized likelihood faster than other block-iterative algorithms which are designed for nonnegatively constrained penalized reconstruction.

Fast iterative image reconstruction methods for fully 3D multispectral bioluminescence tomography

We investigate fast iterative image reconstruction methods for fully 3D multispectral bioluminescence tomography for applications in small animal imaging. Our forward model uses a diffusion approximation for optically inhomogeneous tissue, which we solve using a finite element method (FEM). We examine two approaches to incorporating the forward model into the solution of the inverse problem. In a conventional direct calculation approach one computes the full forward model by repeated solution of the FEM problem, once for each potential source location. We describe an alternative on-the-fly approach where one does not explicitly solve for the full forward model. Instead, the solution to the forward problem is included implicitly in the formulation of the inverse problem, and the FEM problem is solved at each iteration for the current image estimate. We evaluate the convergence speeds of several representative iterative algorithms. We compare the computation cost of those two approaches, concluding that the on-the-fly approach can lead to substantial reductions in total cost when combined with a rapidly converging iterative algorithm.

A three-dimensional statistical approach to improved image quality for multislice helical CT

Multislice helical computed tomography scanning offers the advantages of faster acquisition and wide organ coverage for routine clinical diagnostic purposes. However, image reconstruction is faced with the challenges of three-dimensional cone-beam geometry, data completeness issues, and low dosage. Of all available reconstruction methods, statistical iterative reconstruction (IR) techniques appear particularly promising since they provide the flexibility of accurate physical noise modeling and geometric system description. In this paper, we present the application of Bayesian iterative algorithms to real 3D multislice helical data to demonstrate significant image quality improvement over conventional techniques. We also introduce a novel prior distribution designed to provide flexibility in its parameters to fine-tune image quality. Specifically, enhanced image resolution and lower noise have been achieved, concurrently with the reduction of helical cone-beam artifacts, as demonstrated by phantom studies. Clinical results also illustrate the capabilities of the algorithm on real patient data. Although computational load remains a significant challenge for practical development, superior image quality combined with advancements in computing technology make IR techniques a legitimate candidate for future clinical applications.

Statistical image reconstruction from correlated data with applications to PET

Most statistical reconstruction methods for emission tomography are designed for data modeled as conditionally independent Poisson variates. In reality, due to scanner detectors, electronics and data processing, correlations are introduced into the data resulting in dependent variates. In general, these correlations are ignored because they are difficult to measure and lead to computationally challenging statistical reconstruction algorithms. This work addresses the second concern, seeking to simplify the reconstruction of correlated data and provide a more precise image estimate than the conventional independent methods. In general, correlated variates have a large non-diagonal covariance matrix that is computationally challenging to use as a weighting term in a reconstruction algorithm. This work proposes two methods to simplify the use of a non-diagonal covariance matrix as the weighting term by (a) limiting the number of dimensions in which the correlations are modeled and (b) adopting flexible, yet computationally tractable, models for correlation structure. We apply and test these methods with simple simulated PET data and data processed with the Fourier rebinning algorithm which include the one-dimensional correlations in the axial direction and the two-dimensional correlations in the transaxial directions. The methods are incorporated into a penalized weighted least-squares 2D reconstruction and compared with a conventional maximum a posteriori approach.

An overview of fast convergent ordered-subsets reconstruction methods for emission tomography based on the incremental EM algorithm

Statistical reconstruction has become popular in emission computed tomography but suffers slow convergence (to the MAP or ML solution). Methods proposed to address this problem include the fast but non-convergent OSEM and the convergent RAMLA [1] for the ML case, and the convergent BSREM [2], relaxed OS-SPS and modified BSREM [3] for the MAP case. The convergent algorithms required a user-determined relaxation schedule. We proposed fast convergent OS reconstruction algorithms for both ML and MAP cases, called COSEM (Complete-data OSEM), which avoid the use of a relaxation schedule while maintaining convergence. COSEM is a form of incremental EM algorithm. Here, we provide a derivation of our COSEM algorithms and demonstrate COSEM using simulations. At early iterations, COSEM-ML is typically slower than RAMLA, and COSEM-MAP is typically slower than optimized BSREM while remaining much faster than conventional MAP-EM. We discuss how COSEM may be modified to overcome these limitations.

Improved quantitation for PET/CT image reconstruction with system modeling and anatomical priors

Accurate quantitation of positron emission tomography (PET) tracer uptake levels in tumors is important for staging and monitoring response to treatment. Quantitative accuracy in PET is particularly poor for small tumors because of system partial volume errors and smoothing operations. This work proposes a reconstruction algorithm to reduce the quantitative errors due to limited system resolution and due to necessary image noise reduction. We propose a method for finding and using the detection system response in the projection matrix of a statistical reconstruction algorithm. In addition, we use aligned anatomical information, available in PET/CT scanners, to govern the penalty term applied during each image update. These improvements are combined with Fourier rebinning in a clinically feasible algorithm for reconstructing fully three-dimensional PET data. Results from simulation and measured studies show improved quantitation of tumor values in terms of bias and variance across multiple tumor sizes and activity levels with the proposed method. At common clinical image noise levels for the detection task, the proposed method reduces the error in maximum tumor values by 11% compared to filtered back-projection and 5% compared to conventional iterative methods.

Fusion of Hidden Markov Random Field models and its Bayesian estimation

In this paper, we present a Hidden Markov Random Field (HMRF) data-fusion model. The proposed model is applied to the segmentation of natural images based on the fusion of colors and textons into Julesz ensembles. The corresponding Exploration/ Selection/Estimation (ESE) procedure for the estimation of the parameters is presented. This method achieves the estimation of the parameters of the Gaussian kernels, the mixture proportions, the region labels, the number of regions, and the Markov hyper-parameter. Meanwhile, we present a new proof of the asymptotic convergence of the ESE procedure, based on original finite time bounds for the rate of convergence.

Multigrid tomographic inversion with variable resolution data and image spaces

A multigrid inversion approach that uses variable resolutions of both the data space and the image space is proposed. Since the computational complexity of inverse problems typically increases with a larger number of unknown image pixels and a larger number of measurements, the proposed algorithm further reduces the computation relative to conventional multigrid approaches, which change only the image space resolution at coarse scales. The advantage is particularly important for data-rich applications, where data resolutions may differ for different scales. Applications of the approach to Bayesian reconstruction algorithms in transmission and emission tomography with a generalized Gaussian Markov random field image prior are presented, both with a Poisson noise model and with a quadratic data term. Simulation results indicate that the proposed multigrid approach results in significant improvement in convergence speed compared to the fixed-grid iterative coordinate descent method and a multigrid method with fixed-data resolution.

Penalized maximum-likelihood image reconstruction for lesion detection

Detecting cancerous lesions is one major application in emission tomography. In this paper, we study penalized maximum-likelihood image reconstruction for this important clinical task. Compared to analytical reconstruction methods, statistical approaches can improve the image quality by accurately modelling the photon detection process and measurement noise in imaging systems. To explore the full potential of penalized maximum-likelihood image reconstruction for lesion detection, we derived simplified theoretical expressions that allow fast evaluation of the detectability of a random lesion. The theoretical results are used to design the regularization parameters to improve lesion detectability. We conducted computer-based Monte Carlo simulations to compare the proposed penalty function, conventional penalty function, and a penalty function for isotropic point spread function. The lesion detectability is measured by a channelized Hotelling observer. The results show that the proposed penalty function outperforms the other penalty functions for lesion detection. The relative improvement is dependent on the size of the lesion. However, we found that the penalty function optimized for a 5 mm lesion still outperforms the other two penalty functions for detecting a 14 mm lesion. Therefore, it is feasible to use the penalty function designed for small lesions in image reconstruction, because detection of large lesions is relatively easy.

Iterative reconstruction techniques in emission computed tomography

In emission tomography statistically based iterative methods can improve image quality relative to analytic image reconstruction through more accurate physical and statistical modelling of high-energy photon production and detection processes. Continued exponential improvements in computing power, coupled with the development of fast algorithms, have made routine use of iterative techniques practical, resulting in their increasing popularity in both clinical and research environments. Here we review recent progress in developing statistically based iterative techniques for emission computed tomography. We describe the different formulations of the emission image reconstruction problem and their properties. We then describe the numerical algorithms that are used for optimizing these functions and illustrate their behaviour using small scale simulations.