scispace - formally typeset
Search or ask a question

Showing papers on "Parametric Image published in 2019"


Journal ArticleDOI
TL;DR: This paper first forms a point-source localization (PSL) approach as an optimization problem, then extends and generalizes this formulation to an iterative algorithm, called additive PSL (APSL), for sparse parametric image reconstruction, finding improved image accuracy and computational efficiency with APSL over traditional grid-based approaches.
Abstract: Gamma-ray imaging attempts to reconstruct the spatial and intensity distribution of gamma-emitting radionuclides from a set of measurements. Generally, this problem is solved by discretizing the spatial dimensions and employing the maximum likelihood expectation maximization (ML-EM) algorithm, with or without some form of regularization. While the generality of this formulation enables use in a wide variety of scenarios, it is susceptible to overfitting, limited by the discretization of spatial coordinates, and can be computationally expensive. We present a novel approach to 3D gamma-ray image reconstruction for scenarios where sparsity may be assumed, for example, radiological source search. In this paper, we first formulate a point-source localization (PSL) approach as an optimization problem, where both position and source intensity are continuous variables. We then extend and generalize this formulation to an iterative algorithm, called additive PSL (APSL), for sparse parametric image reconstruction. A set of simulated source search scenarios using a single non-directional detector are considered, finding improved image accuracy and computational efficiency with APSL over traditional grid-based approaches.

26 citations


Proceedings ArticleDOI
01 Mar 2019
TL;DR: Quantification results based on real patient dataset shows that the proposed parametric reconstruction method is better than the Gaussian denoising and non-local mean denoised methods.
Abstract: Deep neural networks have attracted growing interests in medical image due to its success in computer vision tasks. One barrier for the application of deep neural networks is the need of large amounts of prior training pairs, which is not always feasible in clinical practice. Recently, the deep image prior framework shows that the convolutional neural network (CNN) can learn intrinsic structure information from the corrupted image. In this work, an iterative parametric reconstruction framework is proposed using deep neural network as constraint. The network does not need prior training pairs, but only the patient’s own CT image. The training is based on Logan plot derived from multi-bed-position dynamic positron emission tomography (PET) images using 68Ga-PRGD2 tracer. We formulated the estimation of the slope of Logan plot as a constraint optimization problem and solved it using the alternating direction method of multipliers (ADMM) algorithm. Quantification results based on real patient dataset shows that the proposed parametric reconstruction method is better than the Gaussian denoising and non-local mean denoising methods.

15 citations


Journal ArticleDOI
TL;DR: The result showed that the denoised DWI-MRI data obtained using the proposed image denoising method can substantially improve the quality of IVIM parametric images.

9 citations


Journal Article
TL;DR: In this article, the authors demonstrated the capability of total-body parametric imaging using the EXPLORER scanner and showed that the kernel regularized reconstruction and direct reconstruction can achieve superior image quality for tracer kinetics studies compared to the conventional indirect OSEM for total body imaging.
Abstract: 456 Introduction: The EXPLORER consortium has developed the world’s first 2-meter long total-body PET scanner (uEXPLORER) to provide a versatile platform for biomedical research and clinical applications. Its total-body coverage and ultra-high sensitivity provide opportunities for more accurate tracer kinetics analysis in studies of new pharmaceuticals, physiology and pathology. In this work, we demonstrate total-body parametric imaging using kernel and direct reconstruction of the EXPLORER data. Methods: We conducted the first human dynamic total-body PET study using the uEXPLORER scanner. A healthy female subject was recruited (61-yrs old, height 156 cm, weight 56 kg) and gave informed consent under the guidance of the Ethics Board of Zhongshan Hospital (Shanghai, China). A one-hour dynamic scan was performed immediately after an intravenous injection of 6.9 mCi of 18F-FDG in the foot. A total of ~60 billion prompt coincidences were recorded. To exploit the high temporal resolution of the scanner, we divided the dataset into 187 temporal frames: 60×1 sec, 30×2 sec, 20×3 sec, 12×10 sec, 50×30 sec, and 15×120 sec. For quantitative image reconstruction, all corrections (normalization, attenuation, randoms, scatters, and resolution model) were implemented in the forward model. Dynamic data were reconstructed using a 3D TOF list-mode OSEM algorithm and a kernel-based algorithm. We used three composite frames (10-20-30min) to build the kernel matrix. Images were reconstructed into a 239x239x679 matrix with 2.85-mm cubic voxels. The linear Patlak graphical method was analyzed for non-reversible glucose metabolism in different tissues. For comparison, indirect Patlak analysis from reconstructed frames and direct reconstruction using the nested algorithm were conducted for the later 30-min frames. The input function was extracted from the upper aorta region instead of the left ventricle to minimize blurring due to cardiac motion. Results: Images reconstructed by OSEM show good image quality with low noise, even for the 1-second frames. The image quality was further improved by using the kernel method. Total-body Patlak parametric images were obtained by using either indirect estimation or direct reconstruction. In comparison, direct reconstruction improves parametric image quality with better contrast versus background noise tradeoff over the indirect method. Conclusion: This study demonstrated the capability of total-body parametric imaging using the EXPLORER. Furthermore, the results show that the kernel-regularized reconstruction and direct parametric imaging can achieve superior image quality for tracer kinetics studies compared to the conventional indirect OSEM for total-body imaging. Acknowledgements: Support for this work includes NIH grant R01 CA206187. We acknowledge the contributions of all team members from UC Davis, United Imaging Healthcare and Zhongshan Hospital.

7 citations


Journal Article
Tao Feng1, Yizhang Zhao, Yun Dong, Shulin Yao
TL;DR: In this article, a CNN model was trained to predict the images produced using the direct method from the images generated using the indirect approach, and the average normalized RMSD was reduced to 15.5% from 34.5%.
Abstract: 518 Introduction: Parametric imaging using the Patlak model has been shown to provide better quantification and improved specificity for cancer detection compared with SUV imaging. Current methods for generating parametric images include image-based analysis with reconstruction dynamic images (Indirect method) and direct parametric image reconstruction from listmode or sinogram data (Direct method). The direct method has been shown to provide much better image quality compared to the indirect method. However, it has also been shown that the convergence speed for the direct approach is much slower due to the inclusion of the Patlak model into the image reconstruction process. The direct approach also required either the listmode data or the dynamic sinograms, which usually takes a large amount of storage and may not be practical for large-scale analysis. Great potential has been demonstrated using the convolutional neural network (CNN) with deep learning for image processing. In this study, we train a CNN model to predict the images produced using the direct method from the images generated using the indirect approach. Faster reconstruction time and improved image quality can be expected without the need for the listmode or sinogram data. A total of 10 whole body FDG PET scans were collected using the whole-body dynamic scanning protocol, with 2 or 3 dynamic passes and 4 or 5 bed-positions scanned starting 40 minutes post-injection. Every bed position is 4-min long. The study has been approved by the Institutional Review Board of the hospital, and a written informed consent form has been obtained from each patient. The dynamic SUV images were reconstructed using 24 subsets with 2 iterations. Parametric images using the indirect method were acquired by matrix multiplication of the inverse of the Patlak model with the dynamic images. The direct parametric images were generated by combining the Patlak model into the system matrix. A total of 24 subsets and 60 iterations were used in the direct image reconstruction to ensure convergence. The population-based input function was used for both methods. The CNN with 6 layers were used for model training, the Patlak slope and intercept image generated from the indirect method were used as inputs and the slope image generated from the direct method was used as model output. 3D patches with the size of 56[asterisk]56[asterisk]56 were used for both inputs and outputs. A total of 4800 iterations with the batch size being 25 were used for model training. The leave-one-out cross-validation method was applied for the evaluation of the deep learning model. The normalized root mean square difference (RMSD) with the images generated using the direct approach was used for quantitative evaluation criteria. Images predicted using the CNN model (Indirect-CNN) showed much-reduced noise compared with the images generated using the indirect approach, and similar image quality was observed when compared with images generated using the direct approach. The time required for CNN prediction is negligible compared with direct image reconstruction. The average normalized RMSD for images generated using indirect-CNN was reduced to 15.5% from 34.5%, which was the normalized RMSD for images using the indirect method. This study demonstrated the possibility to use CNN to replace the direct parametric image reconstruction process and provide similar parametric image quality using existing SUV images without the requirement of listmode or sinogram data.

1 citations


Journal Article
TL;DR: In this article, a sparsity regularized direct parametric reconstruction algorithm was proposed to suppress the undesirable noise propagation in this ill-posed inverse problem, which collectively uses data from all the dynamic frames while imposing a dictionary learning (DL) based sparsity constraint on K1 spatial variation.
Abstract: 110 Objectives: Dynamic myocardial perfusion imaging (MPI) with PET serves an important role in diagnosis and prognosis of patents with suspected or known coronary artery disease. Conventional myocardial blood flow (MBF) quantification reconstructs a series of dynamic frames and applies a kinetic model to the reconstructed image sequences for measuring the tracer uptake rate K1. This approach leads to noisy K1 estimation due to very limited counts in individual time frames. The goal of this study is to develop a sparsity regularized direct parametric reconstruction algorithm, which collectively uses data from all the dynamic frames while imposing a dictionary learning (DL) based sparsity constraint on K1 spatial variation. Methods: The direct parametric reconstruction is accomplished by relating parametric images to dynamic PET data through a nonlinear transform containing the one-tissue compartment model and the imaging system matrix. To suppress the undesirable noise propagation in this ill-posed inverse problem, we impose a sparsity regularization on the K1 image leading to a penalized log-likelihood function for maximization. The sparsity constraint is constructed as the difference between the estimated K1 image and its sparse representation based on the learned dictionary from a self-created hollow sphere. A two-stage iterative approach is adopted to solve this optimization problem. In stage one, we calculate the sparse representation of the current estimation of K1 given the learned dictionary. In stage two, the penalized log-likelihood function is optimized with the sparsity penalty term fixed. Applying optimization transfer, we construct separable surrogate functions for the log-likelihood term and the sparsity constraint term, respectively. The combined surrogate function, which is solvable by convenient voxel-wise optimization is maximized by the damped Newton method. To evaluate the proposed algorithm, we simulated two sets of realistic Rb-82 dynamic MPI PET data, one with normal MBF (K1=1.48) and the other with regionally reduced MBF (K1=1.13). Using the XCAT phantom, PET image frames were created by assigning the activities integrated from the multiple organ time activity curves based on clinical measurement. We performed analytic simulations for the geometry of a GE RX PET scanner to generate 20 noise realizations for each dynamic dataset. By assessing the ensemble noise versus bias tradeoff of the normal and abnormal K1 on the region of interest, we compared the proposed method, the conventional method with and without post filtering, and a quadratic penalty regularized direct parametric reconstruction method. We also evaluated the tradeoff of the ensemble noise and the contrast between the normal and the defect K1 for abnormal MBF detectability. Results: For the regional normal K1 estimation, the mean and the ensemble normalized standard deviation (EnNSD) across 20 noise realizations obtained by the conventional method without and with post filtering, the quadratic penalty regularized direct algorithm, and the proposed method are 1.43+/-0.45, 1.10+/-0.21, 1.36+/-0.22, and 1.45+/-0.22. For the abnormal case, the corresponding regional mean+/-EnNSD are 1.13+/-0.64, 0.78+/-0.26, 1.02+/-0.27, and 1.07+/-0.27, respectively. In both cases, post filtering in the conventional method reduces noise at the cost of introducing large bias. The proposed sparsity constraint outperforms the quadratic penalty, achieving similar noise but reduced bias resulting in better recovered contrast. Conclusions: We developed a sparsity constrained direct parametric image reconstruction algorithm that incorporates the DL based regularization on the K1 parametric image. Using simulated dynamic MPI PET data, we demonstrated its better performance in K1 estimation and abnormal K1 detection compared with conventional methods. The proposed method shows its potential to advance MBF quantification in dynamic PET MPI.

1 citations



Proceedings ArticleDOI
01 Oct 2019
TL;DR: It is shown that with the use of total-body PET and the increased sensitivity, it is possible to estimate parametric images using the very early stages of the FDG injection, and the combined effects of delay and dispersion will be studied in the future.
Abstract: The recently developed total-body PET scanner enables high temporal resolution in dynamic imaging. Due to the much improved temporal resolution and large field of view, delay and the dispersion effects in the image-derived input function, which vary for different tissues and organs, may affect accuracy in parametric imaging. In this paper, the delay effect was studied using the early kinetics of an FDG scan, which may be approximated using a 1-tissue compartment model. Dynamic reconstructed frames were acquired using the total-body PET scanner with 1-second frames for the first 30 seconds and 2 seconds for the subsequent 60 seconds. The image-derived input function was acquired from the reconstructed dynamic sequence using volumes of interest in the ascending and descending aorta. Voxel-specific delay times for the plasma input function were also modeled within the 1-tissue compartment model. A total of 4 parametric images were generated. Image-based parametric image generation was achieved with a maximum likelihood estimation method. Parametric images with and without the modeling of delay time in the input function were compared. Additional image denoising techniques including Gaussian denoising and non-local-mean denoising were employed. Quantitative evaluation was achieved by the calculation of the Akaike Information Criterion (AIC). The voxel-specific parameters of the 1-tissue compartment together with the delay time were successfully reconstructed using the proposed method. The estimated delay time showed variations as large as 40 seconds. The non-local-mean filter was shown to be able to reduce the image noise of the generated parametric images. Various image artifacts were observed when no delay time model was included. We have shown that with the use of total-body PET and the increased sensitivity, it is possible to estimate parametric images using the very early stages of the FDG injection. The combined effects of delay and dispersion will be studied in the future.

1 citations


Book ChapterDOI
26 Sep 2019
TL;DR: This work uses the changes in backscattered energy (CBE) to improve the visual difference between the muscle and fat tissue, because some studies show that CBEs calculated from RF signals present different behaviors when scattered by particles with different characteristics.
Abstract: Ultrasonic radiation can be used as a non-invasive, ionizing radiation-free, portable and inexpensive imaging technique that allows to acquire images in real time. However, this technique presents some limitations, especially when referring to image quality, for example it is not easy to distinguish some soft tissues. Several methods for improving the visual quality have been developed, which are based, in most of cases, on disturbances of the medium with external stimuli. This work uses the changes in backscattered energy (CBE) to improve the visual difference between the muscle and fat tissue, because some studies show that CBEs calculated from RF signals present different behaviors when scattered by particles with different characteristics. The new image has more visual details when compare with the conventional ultrasound images. Two numerical phantoms were simulated with different kinds of scatterers. Conventional B-mode images were obtained for different temperatures (37 °–40 °C). A new parametric image was proposed by using the angular coefficient of the curve pixel intensity versus temperature, obtained for each pixel. The proposed parametric image was able to enhance the visual contrast between the simulated tissues.

Proceedings ArticleDOI
01 Oct 2019
TL;DR: A penalized direct reconstruction method is proposed in which a penalty is imposed in parametric images for the accurate estimation of parametric image and it is demonstrated that the proposed method can estimate the parametricimage only using 20 min acquisition.
Abstract: Parametric positron emission tomography (PET) imaging can provide more quantitative information compared to the standard uptake value (SUV). To acquire the parametric image, a long dynamic scan over 80 mins are necessary, however, some patients with severe diseases cannot stay in a scanner such a long time or can produce bad motion artifacts during the dynamic scan. To address this issue, we propose a novel parametric PET reconstruction method using a dual injection strategy. The dual injection strategy is that the first injection of half dose at the beginning and the second injection of half dose at the middle of scan are administered, which can preserve total amount of dose. A scan is conducted after 60 mins from the first injection with a duration of total 20 mins as usually done in PET scan protocol. To estimate the parametric image, we firstly split the first and last time activity curves (TACs), and then a parametric image is estimated using partial dynamic data based on the simplified reference tissue model (SRTM). To split the TACs, a linear fitting method is used by assuming that there is no temporal metabolite changes and TAC with 20 min in the steady-state is linear. For the accurate estimation of parametric image, a penalized direct reconstruction method is proposed in which a penalty is imposed in parametric images. In brain phantom simulation, we demonstrated that the proposed method can estimate the parametric image only using 20 min acquisition.

Posted Content
TL;DR: Novel methods that incorporate both 2D spatial and 1D temporal penalties produced dynamic PET images of higher quality than conventional 2D methods, w/o need for post-filtering.
Abstract: Our aim was to enhance visual quality and quantitative accuracy of dynamic positron emission tomography (PET)uptake images by improved image reconstruction, using sophisticated sparse penalty models that incorporate both 2D spatial+1D temporal (3DT) information. We developed two new 3DT PET reconstruction algorithms, incorporating different temporal and spatial penalties based on discrete cosine transform (DCT)w/ patches, and tensor nuclear norm (TNN) w/ patches, and compared to frame-by-frame methods; conventional 2D ordered subsets expectation maximization (OSEM) w/ post-filtering and 2D-DCT and 2D-TNN. A 3DT brain phantom with kinetic uptake (2-tissue model), and a moving 3DT cardiac/lung phantom was simulated and reconstructed. For the cardiac/lung phantom, an additional cardiac gated 2D-OSEM set was reconstructed. The structural similarity index (SSIM) and relative root mean squared error (rRMSE) relative ground truth was investigated. The image derived left ventricular (LV) volume for the cardiac/lung images was found by region growing and parametric images of the brain phantom were calculated. For the cardiac/lung phantom, 3DT-TNN yielded optimal images, and 3DT-DCT was best for the brain phantom. The optimal LV volume from the 3DT-TNN images was on average 11 and 55 percentage points closer to the true value compared to cardiac gated 2D-OSEM and 2D-OSEM respectively. Compared to 2D-OSEM, parametric images based on 3DT-DCT images generally had smaller bias and higher SSIM. Our novel methods that incorporate both 2D spatial and 1D temporal penalties produced dynamic PET images of higher quality than conventional 2D methods, w/o need for post-filtering. Breathing and cardiac motion were simultaneously captured w/o need for respiratory or cardiac gating. LV volumes were better recovered, and subsequently fitted parametric images were generally less biased and of higher quality.