Enhanced 3D PET OSEM reconstruction using inter-update Metz filtering

Deep learning-assisted PET imaging achieves fast scan/low-dose examination

PURPOSE

This study aimed to investigate the impact of a deep learning (DL)-based denoising method on the image quality and lesion detectability of ¹⁸F-FDG positron emission tomography (PET) images.

METHODS

Fifty-two oncological patients undergoing an ¹⁸F-FDG PET/CT imaging with an acquisition of 180 s per bed position were retrospectively included. The list-mode data were rebinned into four datasets: 100% (reference), 75%, 50%, and 33.3% of the total counts, and then reconstructed by OSEM algorithm and post-processed with the DL and Gaussian filter (GS). The image quality was assessed using a 5-point Likert scale, and FDG-avid lesions were counted to measure lesion detectability. Standardized uptake values (SUVs) in livers and lesions, liver signal-to-noise ratio (SNR) and target-to-background ratio (TBR) values were compared between the methods. Subgroup analyses compared TBRs after categorizing lesions based on parameters like lesion diameter, uptake or patient habitus.

RESULTS

The DL method showed superior performance regarding image noise and inferior performance regarding lesion contrast in the qualitative assessment. More than 96.8% of the lesions were successfully identified in DL images. Excellent agreements on SUV in livers and lesions were found. The DL method significantly improved the liver SNR for count reduction down to 33.3% (p < 0.001). Lesion TBR was not significantly different between DL and reference images of the 75% dataset; furthermore, there was no significant difference either for lesions of > 10 mm or lesions in BMIs of > 25. For the 50% dataset, there was no significant difference between DL and reference images for TBR of lesion with > 15 mm or higher uptake than liver.

CONCLUSIONS

The developed DL method improved both liver SNR and lesion TBR indicating better image quality and lesion conspicuousness compared to GS method. Compared with the reference, it showed non-inferior image quality with reduced counts by 25-50% under various conditions.

MR-Guided Delivery of Hydrophilic Molecular Imaging Agents Across the Blood-Brain Barrier Through Focused Ultrasound

PURPOSE

A wide variety of hydrophilic imaging and therapeutic agents are unable to gain access to the central nervous system (CNS) due to the blood-brain barrier (BBB). In particular, unless a particular transporter exists that may transport the agent across the BBB, most agents that are larger than 500 Da or that are hydrophilic will be excluded by the BBB. Glutamate carboxypeptidase II (GCPII), also known as the prostate-specific membrane antigen (PSMA) in the periphery, has been implicated in various neuropsychiatric conditions. As all agents that target GCPII are hydrophilic and thereby excluded from the CNS, we used GCPII as a platform for demonstrating our MR-guided focused ultrasound (MRgFUS) technique for delivery of GCPII/PSMA-specific imaging agents to the brain.

PROCEDURES

Female rats underwent MRgFUS-mediated opening of the BBB. After opening of the BBB, either a radio- or fluorescently labeled ureido-based ligand for GCPII/PSMA was administered intravenously. Brain uptake was assessed for 2-(3-{1-carboxy-5-[(6-[¹⁸F]fluoro-pyridine-3-carbonyl)-amino]-pentyl}-ureido)-pentanedioic acid ([¹⁸F]DCFPyL) and YC-27, two compounds known to bind GCPII/PSMA with high affinity, using positron emission tomography (PET) and near-infrared fluorescence (NIRF) imaging, respectively. Specificity of ligand binding to GCPII/PSMA in the brain was determined with co-administration of a molar excess of ZJ-43, a compound of the same chemical class but different structure from either [¹⁸F]DCFPyL or YC-27, which competes for GCPII/PSMA binding.

RESULTS

Dynamic PET imaging using [¹⁸F]DCFPyL demonstrated that target uptake reached a plateau by ∼1 h after radiotracer administration, with target/background ratios continuing to increase throughout the course of imaging, from a ratio of ∼4:1 at 45 min to ∼7:1 by 80 min. NIRF imaging likewise demonstrated delivery of YC-27 to the brain, with clear visualization of tracer in the brain at 24 h. Tissue uptake of both ligands was greatly diminished by ZJ-43 co-administration, establishing specificity of binding of each to GCPII/PSMA. On gross and histological examination, animals showed no evidence for hemorrhage or other deleterious consequences of MRgFUS.

CONCLUSIONS

MRgFUS provided safe opening of the BBB to enable specific delivery of two hydrophilic agents to target tissues within the brain. This platform might facilitate imaging and therapy using a variety of agents that have heretofore been excluded from the CNS.

Development of Radiolabeled Ligands Targeting the Glutamate Binding Site of the N-Methyl-d-aspartate Receptor as Potential Imaging Agents for Brain

Abnormal activity of various N-methyl-d-aspartate receptor (NMDAR) subtypes has been implicated in a wide variety of neurological disorders such as Alzheimer's disease, schizophrenia, and epilepsy. Imaging agents for PET and SPECT that target NMDARs in a subtype-selective fashion may enable better characterization of those disorders and enhance drug development. On the basis of a pyrazoline derivative that demonstrated neuroprotective effects in vivo, we synthesized a series of para-substituted analogues and measured their affinities to various NMDAR subtypes. Compounds 4a-c and 4e showed greater, nanomolar affinity for the GluN1/2A subtype versus GluN1/2B. Dicarbomethoxy (pro-drug) analogues of [^124/125I]4d and [¹¹C]4e (i.e., [^124/125I]11d and [¹¹C]11e) were generated and tested for NMDAR binding specificity in ex vivo autoradiography and brain biodistribution studies. Although NMDAR-specific binding could be demonstrated for [¹²⁵I]11d and [¹¹C]11e through autoradiography and biodistribution studies, imaging of neither [¹²⁴I]11d nor [¹¹C]11e could demonstrate brain penetration sufficient for detection by PET.

Integration of advanced 3D SPECT modeling into the open-source STIR framework

PURPOSE

The Software for Tomographic Image Reconstruction (STIR, http://stir.sourceforge.net) package is an open source object-oriented library implemented in C++. Although its modular design is suitable for reconstructing data from several modalities, it currently only supports Positron Emission Tomography (PET) data. In this work, the authors present results for Single Photon Emission Computed Tomography (SPECT) imaging.

METHODS

This was achieved by the complete integration of a 3D SPECT system matrix modeling library into STIR.

RESULTS

The authors demonstrate the flexibility of the combined software by reconstructing simulated and acquired projections from three different scanners with different iterative algorithms of STIR.

CONCLUSIONS

The extension of the open source STIR project with advanced SPECT modeling will enable the research community to study the performance of several algorithms on SPECT data, and potentially implement new algorithms by expanding the existing framework.

Four-Dimensional Image Reconstruction Strategies in Cardiac-Gated and Respiratory-Gated PET Imaging

Cardiac and respiratory movements pose significant challenges to image quality and quantitative accuracy in PET imaging. Cardiac and/or respiratory gating attempt to address this issue, but instead lead to enhanced noise levels. Direct four-dimensional (4D) PET image reconstruction incorporating motion compensation has the potential to minimize noise amplification while removing considerable motion blur. A wide-ranging choice of such techniques is reviewed in this work. Future opportunities and the challenges facing the adoption of 4D PET reconstruction and its role in basic and clinical research are also discussed.

Automated analysis of small animal PET studies through deformable registration to an atlas

Purpose

This work aims to develop a methodology for automated atlas-guided analysis of small animal positron emission tomography (PET) data through deformable registration to an anatomical mouse model.

Methods

A non-rigid registration technique is used to put into correspondence relevant anatomical regions of rodent CT images from combined PET/CT studies to corresponding CT images of the Digimouse anatomical mouse model. The latter provides a pre-segmented atlas consisting of 21 anatomical regions suitable for automated quantitative analysis. Image registration is performed using a package based on the Insight Toolkit allowing the implementation of various image registration algorithms. The optimal parameters obtained for deformable registration were applied to simulated and experimental mouse PET/CT studies. The accuracy of the image registration procedure was assessed by segmenting mouse CT images into seven regions: brain, lungs, heart, kidneys, bladder, skeleton and the rest of the body. This was accomplished prior to image registration using a semi-automated algorithm. Each mouse segmentation was transformed using the parameters obtained during CT to CT image registration. The resulting segmentation was compared with the original Digimouse atlas to quantify image registration accuracy using established metrics such as the Dice coefficient and Hausdorff distance. PET images were then transformed using the same technique and automated quantitative analysis of tracer uptake performed.

Results

The Dice coefficient and Hausdorff distance show fair to excellent agreement and a mean registration mismatch distance of about 6 mm. The results demonstrate good quantification accuracy in most of the regions, especially the brain, but not in the bladder, as expected. Normalized mean activity estimates were preserved between the reference and automated quantification techniques with relative errors below 10 % in most of the organs considered.

Conclusion

The proposed automated quantification technique is reliable, robust and suitable for fast quantification of preclinical PET data in large serial studies.

Convergence optimization of parametric MLEM reconstruction for estimation of Patlak plot parameters

In dynamic positron emission tomography data many researchers have attempted to exploit kinetic models within reconstruction such that parametric images are estimated directly from measurements. This work studies a direct parametric maximum likelihood expectation maximization algorithm applied to [(18)F]DOPA data using reference-tissue input function. We use a modified version for direct reconstruction with a gradually descending scheme of subsets (i.e. 18-6-1) initialized with the FBP parametric image for faster convergence and higher accuracy. The results compared with analytic reconstructions show quantitative robustness (i.e. minimal bias) and clinical reproducibility within six human acquisitions in the region of clinical interest. Bland-Altman plots for all the studies showed sufficient quantitative agreement between the direct reconstructed parametric maps and the indirect FBP (--0.035x+0.48E--5).

From 2D PET to 3D PET: Issues of Data Representation and Image Reconstruction

Positron emission tomography (PET), intrinsically a 3D imaging technique, was for a long time exclusively operated in 2D mode, using septa to shield the detectors from photons emitted obliquely to the detector planes. However, the use of septa results in a considerable loss of sensitivity. From the late 1980s, significant efforts have been made to develop a methodology for the acquisition and reconstruction of 3D PET data. This paper focuses on the differences between data acquisition in 2D and 3D mode, especially in terms of data set sizes and representation. Although the real time data acquisition aspect in 3D has been mostly solved in modern PET scanner systems, there still remain questions on how to represent and how to make best use of the information contained in the acquired data sets. Data representation methods, such as list-mode and matrix-based methods, possibly with additional compression, will be discussed. Moving from 2D to 3D PET has major implications on the way these data are reconstructed to images. Two fundamentally different approaches exist, the analytical one and the iterative one. Both, at different expenses, can be extended to directly handle 3D data sets. Either way the computational burden increases heavily compared to 2D reconstruction. One possibility to benefit from the increased sensitivity in 3D PET while sticking to high-performance 2D reconstruction algorithms is to rebin 3D into 2D data sets. The value of data rebinning will be explored. An ever increasing computing power and the concept of distributed or parallel computing have made direct 3D reconstruction feasible. Following a short review of reconstruction methods and their extensions to 3D, we focus on numerical aspects that improve reconstruction performance, which is especially important in solving large equation systems in 3D iterative reconstruction. Finally exemplary results are shown to review the properties of the discussed algorithms. This paper concludes with an overview on future trends in data representation and reconstruction.

Object dependency of resolution in reconstruction algorithms with interiteration filtering applied to PET data

In this paper, we study the resolution properties of those algorithms where a filtering step is applied after every iteration. As concrete examples we take filtered preconditioned gradient descent algorithms for the Poisson log likelihood for PET emission data. For nonlinear estimators, resolution can be characterized in terms of the linearized local impulse response (LLIR). We provide analytic approximations for the LLIR for the class of algorithms mentioned above. Our expressions clearly show that when interiteration filtering (with linear filters) is used, the resolution properties are, in most cases, spatially varying, object dependent and asymmetric. These nonuniformities are solely due to the interaction between the filtering step and the Poisson noise model. This situation is similar to penalized likelihood reconstructions as studied previously in the literature. In contrast, nonregularized and postfiltered maximum-likelihood expectation maximization (MLEM) produce images with nearly "perfect" uniform resolution when convergence is reached. We use the analytic expressions for the LLIR to propose three different approaches to obtain nearly object independent and uniform resolution. Two of them are based on calculating filter coefficients on a pixel basis, whereas the third one chooses an appropriate preconditioner. These three approaches are tested on simulated data for the filtered MLEM algorithm or the filtered separable paraboloidal surrogates algorithm. The evaluation confirms that images obtained using our proposed regularization methods have nearly object independent and uniform resolution.

A unified noise analysis for iterative image estimation

Iterative image estimation methods have been widely used in emission tomography. Accurate estimation of the uncertainty of the reconstructed images is essential for quantitative applications. While both iteration-based noise analysis and fixed-point noise analysis have been developed, current iteration-based results are limited to only a few algorithms that have an explicit multiplicative update equation and some may not converge to the fixed-point result. This paper presents a theoretical noise analysis that is applicable to a wide range of preconditioned gradient-type algorithms. Under a certain condition, the proposed method does not require an explicit expression of the preconditioner. By deriving the fixed-point expression from the iteration-based result, we show that the proposed iteration-based noise analysis is consistent with fixed-point analysis. Examples in emission tomography and transmission tomography are shown. The results are validated using Monte Carlo simulations.

A prospective study of 2-[18F] fluoro-2-deoxy-D-glucose/positron emission tomography scan, 99mTc-labeled arcitumomab (CEA-scan), and blind second-look laparotomy for detecting colon cancer recurrence in patients with increasing carcinoembryonic antigen levels

BACKGROUND

An increasing carcinoembryonic antigen (CEA) level in the absence of disease on imaging studies can present a diagnostic challenge. We evaluated 2-[18F] fluoro-2-deoxy-D-glucose and positron emission tomography (FDG-PET) scan and CEA scan before second-look laparotomy as a means of localizing recurrent colorectal cancer.

METHODS

Patients underwent computed tomography scan, bone scan, colonoscopy, and magnetic resonance imaging, and those without evidence of disease or resectable disease in the abdomen had FDG-PET and CEA scans. At second-look laparotomy, a surgeon blinded to the results of the FDG-PET and CEA scans performed an exploration and mapped findings. A second surgeon, with knowledge of the FDG-PET and CEA scans, then explored the patient; all lesions were biopsied or resected for pathology.

RESULTS

In 28 patients explored, disease was found at operation in 26 (94%). Ten had unresectable disease. FDG-PET scans predicted unresectable disease in 90% of patients. CEA scans failed to predict unresectable disease in any patient. In 16 patients found to have resectable disease or disease that could be treated with regional therapy, FDG-PET scan predicted this in 81% and CEA scan in 13%.

CONCLUSIONS

FDG-PET scan can predict those patients who would likely benefit from a laparotomy. If the FDG-PET scan indicates resectable disease, laparotomy can be considered. However, if the findings predict unresectable disease or the absence of disease, the patient should pursue systemic therapy or continued observation.