M6A1  Parametric Imaging and Tracer Kinetic Modeling

Saturday, Nov. 7  08:30-10:00  Golden Pacific Ballroom

Session Chair:  quanzheng li, MGH, Harvard Medical School, United States; Richard Carson, Yale University, United States

Show/Hide All Abstracts

(08:30) M6A1-1, Direct Reconstruction Using Partially Linearized Particle Swarm Optimization Penalized by Non Local Means Prior

S. K. Kang, S. Seo, J. S. Lee

Seoul National University, Seoul, Koera
Kinetic analysis of the dynamic PET scan data based on compartmental modeling allows us to quantify the rate constants of radiotracer exchange between the compartments. However, tomographic reconstruction performed prior to the kinetic analysis makes the exact noise modeling in the kinetic analysis difficult, leading to noisy parametric images. Accordingly, the efforts have been made to combine the image reconstruction and kinetic analysis (direct reconstruction). However, conventional optimization methods, such as Levenberg-Marquardt Algorithm (LMA) is not appropriate to find the solution of direct reconstruction due to the non-convexity and non-differentiability of its cost function. Therefore, we propose a direct reconstruction method using particle swam optimization (PSO) to resolve this problem because PSO searches optimal candidate based on pre-calculated particle sets. In this study, we developed a penalized likelihood framework based on this new algorithm and compared its performance of this new algorithm with the conventional LMA method using simulation data. Our new direct reconstruction method employs partially linearized SPO (PL-PSO) and non-local mean (NLM) smoothing prior, and remarkably reduced the image variance relative to the indirect estimation. In addition, the incorporation of NLM prior leaded to the better quantitative accuracy in comparison with the only use of typical local smoothing prior except for K1 estimation.

(08:45) M6A1-2, Theoretical Analysis of Lesion Quantification in Penalized Maximum-Likelihood Patlak Parametric Image Reconstruction for Dynamic PET

L. Yang, J. Qi

Dept. of Biomedical Engineering, University of California Davis, Davis, CA, USA
Quantification of tracer kinetics is one of the important tasks in molecular imaging using dynamic positron emission tomography (PET). One popular method to analyze dynamic PET data is the Patlak graphical model. The conventional method to generate Patlak parametric images is to reconstruct a sequence of dynamic images first and then perform the Patlak analysis pixel-by-pixel, which we refer to as the indirect method. Alternatively, direct reconstruction methods estimate Patlak parametric images directly from dynamic sinogram data by incorporating the Patlak model into the image reconstruction procedure. In this work, we theoretically analyze the performance of lesion quantification for penalized maximum-likelihood (PML) image reconstruction in both the indirect and direct methods. Simplified expressions for evaluating the bias, variance, and ensemble mean squared error (EMSE) of the estimated Patlak parameters have been derived and applied to guide the selection of the regularization parameters to minimize the EMSE of lesion quantification in the Patlak slope image. The proposed method is validated using computer-based Monte Carlo simulations. Good agreements between theoretical predictions and Monte Carlo results are observed.

(09:00) M6A1-3, Direct Estimation of Neurotransmitter Response in Awake and Freely Moving Animals

G. I. Angelis1, J. E. Gillam1, W. J. Ryder1, A. Z. Kyme1,2, R. R. Fulton1,3,4, S. R. Meikle1

1Faculty of Health Sciences, Brain and Mind Research Institute, The University of Sydney, Sydney, Australia
2Biomedical Engineering, UC Davis, Davis, CA
3Department of Medical Physics, Westmead Hospital, Sydney, Australia
4School of Physics, The University of Sydney, Sydney, Australia
The temporal characterisation of endogenous neurotransmitter release during a cognitive task or drug intervention is an important capability for studying the role of neurotransmitters in normal and aberrant brain function, including disease. Advanced kinetic models, such as the linear parametric neurotransmitter PET (lp-ntPET) have been developed to appropriately model the transient changes in the model parameters, such as the radiotracer efflux from the target tissue, during endogenous neurotransmitter release. Incorporation of the kinetic model within the tomographic reconstruction algorithm may lead to improved parameter estimates, both in terms of precision and accuracy, compared to the conventional two-step post reconstruction approach. In this study, we evaluate a direct reconstruction approach that uses an expectation maximisation framework to transfer the 4D spatiotemporal maximum likelihood problem into an image-based weighted least squares problem. This framework allows the use of well established kinetic models, such as the lp-ntPET model, to estimate the endogenous neurotransmitter response directly from the dynamic PET data. Dynamic GATE simulations using a realistic digital rat brain phantom showed that the proposed direct reconstruction method can provide higher temporal accuracy and precision for the estimated neurotransmitter response at the voxel level, compared to the conventional post reconstruction modelling. In addition, we applied the developed methodology to a [11C]raclorpide displacement study on an awake and freely moving rat and generated voxel-wise parametric maps illustrating ligand displacement from striatum.

(09:15) M6A1-4, Improved Poisson-Guided Clustering of Dynamic PET Imaging Data to Extract Shape Patterns for Mixture Modelling and Parametric Imaging.

F. R. Hernandez Fernandez, F. O'Sullivan

Statistics, University College Cork, Cork, Ireland
Segmentation and clustering are increasingly used as part of the analysis procedures applied to imaging data from dynamic PET studies. While well-established methods as hierarchical clustering and K-means algorithms have been found useful, there is interest in investigating if methods that attempt to directly extract shape patterns and make use of the Poisson character of the data might be worthwhile. Considering PET signals are independent time series over the scan duration with Poisson structure of mean µi(t)?-1 for some overdispersion noise-factor ? then we can construct a likelihood-based deviance criterion for assessment of the shape means. If we can identify K different shape-patterns then each signal can be characterised by a combination of K clusters with an intensity associated to that signal. Knowledge of the signal provides estimation of the intensity equivalent to IRLS and the deviance objective function can be manipulated for the shape optimisation (µ¯k). The criterion can be embedded in hierarchical algorithms to produce a deviance-guided recursive clustering method. Knowledge of the noise is used to construct an information criterion for assessment of the number of cluster. We evaluate the deviance guided clustering procedure in comparison to K-means in the context of Verapamil studies [Deo et al, 2014]. Image segmentation was performed on the PET data and regional comparison of the fitted signals from the two approaches carried out. Simulations over a range of noise levels and as a function of the homogeneity of the clustering were also evaluated. Adaptation of multivariate clustering to facilitate extraction of shape patterns in pseudo-Poisson data is easily accomplished and beneficial. The approach has potential in the context of the kinetic analysis of PET data where the empirical support for the pseudo-Poisson approximation is more compelling than the Gaussian assumptions that underlie K-means clustering.

(09:30) M6A1-5, Assessment of Kinetic Modeling Quality of Fit by Cluster Analysis of Residuals: Application to Direct Reconstruction of Cardiac PET Data

M. Germino1, A. J. Sinusas2, C. Liu1, R. E. Carson1

1Dept. of Biomedical Engineering, Yale University, New Haven, CT, USA
2Dept. of Internal Medicine and Diagnostic Radiology, Yale University, New Haven, CT, USA
Direct reconstruction algorithms for PET are capable of producing lower variance parametric images than those from the traditional frame-based approach. However, if the selected kinetic model is not appropriate for all voxels in the field-of-view, structured residuals from poor model fits can propagate into regions-of-interest where the model is accurate. We previously proposed a direct reconstruction algorithm for cardiac PET that fits the kinetic model only at voxels in a pre-specified mask, and uses cubic B-splines elsewhere. Here we present an automated approach to generate this mask, based on k-means clustering of the residuals from a frame-based fit. In this application, the cluster-based mask allowed the kinetic model to be fit to many more voxels than those defined by a manually drawn myocardium region. However, the resulting parameter estimates had comparable agreement with frame-based estimates as those generated using the more restrictive mask. The hybrid method with either the manually defined or cluster-based mask came closer to the frame-based method than a direct reconstruction using the kinetic model in all voxels, suggesting that the hybrid method with a cluster-based mask can reduce bias from error propagation of non-model fits.

(09:45) M6A1-6, Whole Body Parametric Imaging on Clinical Scanner: Direct 4D Reconstruction with Simultaneous Attenuation Estimation and Time-Dependent Normalization

V. Y. Panin1, H. Bal1, M. Defrise1, M. E. Casey1, N. A. Karakatsanis2, A. Rahmim3

1Molecular Imaging, Siemens Healthcare, Knoxville, TN, USA
2the Icahn School of Medicine, Mt Sinai Health System, New York, NY, USA
3Departments of Radiology and Electrical & Computer Engineering, Johns Hopkins University, Baltimore, MD, USA
Whole body dynamic PET imaging has the potential to enhance detectability and quantification when assessing disease stage or progress. The same body region is repeatedly scanned within relatively short acquisition frames and with delays between time samples. Repeated scanning is sensitive to patient motion, which may cause mismatches between attenuation and activity maps and thus erroneous correction factors. Moreover, since count rates change with time, standard software correction factors can be time dependent. The generation of parametric images requires proper physiological modeling and has been shown to benefit from so-called direct 4D reconstruction methods. In this work we extend the MLACF/MLAA algorithms for application to dynamic direct reconstruction. Handling time-dependent normalization requires a redesign of the existing algorithm as well. The reconstruction methodology was verified on Siemens mCT scanner patient data using the standard Patlak model. Different number of frames and scan initiation time points were investigated. Initial results showed that direct 4D reconstruction outperformed the indirect approach. Available CT attenuation information can be corrected based on emission data.