PET image reconstruction using kernel method

The Evolution of Artificial Intelligence in Nuclear Medicine

Nuclear medicine has continuously evolved since its beginnings, constantly improving the diagnosis and treatment of various diseases. The integration of artificial intelligence (AI) is one of the latest revolutionizing chapters, promising significant advancements in diagnosis, prognosis, segmentation, image quality enhancement, and theranostics. Early AI applications in nuclear medicine focused on improving diagnostic accuracy, leveraging machine learning algorithms for disease classification and outcome prediction. Advances in deep learning, including convolutional and more recently transformer-based neural networks, have further enabled more precise diagnosis and image segmentation as well as low-dose imaging, and patient-specific dosimetry for personalized treatment. Generative AI, driven by large language models and diffusion techniques, is now allowing the process, interpretation, and generation of complex medical language and images. Despite these achievements, challenges such as data scarcity, heterogeneity, and ethical concerns remain barriers to clinical translation. Addressing these issues through interdisciplinary collaboration will pave the way for a broader adoption of AI in nuclear medicine, potentially enhancing patient care and optimizing diagnosis and therapeutic outcomes.

Total-Body Parametric Imaging Using Relative Patlak Plot

The standard Patlak plot, a simple yet efficient model, is widely used to describe irreversible tracer kinetics for dynamic PET imaging. Its widespread application to whole-body parametric imaging remains constrained because of the need for a full-time-course input function (e.g., 1 h). In this paper, we demonstrate the relative Patlak (RP) plot, which eliminates the need for the early-time input function, for total-body parametric imaging and its application to 20-min clinical scans acquired in list mode. Methods: We conducted a theoretic analysis to indicate that the RP intercept b' is equivalent to a ratio of the SUV relative to the plasma concentration, whereas the RP slope Ki ' is equal to the standard Patlak Ki (net influx rate) multiplied by a global scaling factor for each subject. One challenge in applying RP to a short scan duration (e.g., 20 min) is the resulting high noise in the parametric images. We applied a self-supervised deep-kernel method for noise reduction. Using the standard Patlak plot as the reference, the RP method was evaluated for lesion quantification, lesion-to-background contrast, and myocardial visualization in total-body parametric imaging in 22 human subjects (12 healthy subjects and 10 cancer patients) who underwent a 1-h dynamic 18F-FDG scan. The RP method was also applied to the dynamic data reconstructed from a clinical standard 20-min list-mode scan either at 1 or 2 h after injection for 2 cancer patients. Results: We demonstrated that it is feasible to obtain high-quality parametric images from 20-min scans using RP parametric imaging with a self-supervised deep-kernel noise-reduction strategy. The RP slope Ki ' was highly correlated with the standard Patlak Ki in lesions and major organs, demonstrating its quantitative potential across subjects. Compared with conventional SUVs, the Ki ' images significantly improved lesion contrast and enabled visualization of the myocardium for potential cardiac assessment. The application of the RP parametric imaging to the 2 clinical scans also showed similar benefits. Conclusion: Using total-body PET with the RP approach, it is feasible to generate parametric images using data from a 20-min clinical list-mode scan.

Quantitative PET imaging and modeling of molecular blood-brain barrier permeability

Neuroimaging of blood-brain barrier permeability has been instrumental in identifying its broad involvement in neurological and systemic diseases. However, current methods evaluate the blood-brain barrier mainly as a structural barrier. Here we developed a non-invasive positron emission tomography method in humans to measure the blood-brain barrier permeability of molecular radiotracers that cross the blood-brain barrier through its molecule-specific transport mechanism. Our method uses high-temporal resolution dynamic imaging and kinetic modeling for multiparametric imaging and quantification of the blood-brain barrier permeability-surface area product of molecular radiotracers. We show, in humans, our method can resolve blood-brain barrier permeability across three radiotracers and demonstrate its utility in studying brain aging and brain-body interactions in metabolic dysfunction-associated steatotic liver inflammation. Our method opens new directions to effectively study the molecular permeability of the human blood-brain barrier in vivo using the large catalogue of available molecular positron emission tomography tracers.

Imaging Demyelinated Axons After Spinal Cord Injuries with PET Tracer [18F]3F4AP

Spinal cord injuries (SCIs) often lead to lifelong disability. Among the various types of injuries, incomplete and discomplete injuries, where some axons remain intact, offer potential for recovery. However, demyelination of these spared axons can worsen disability. Demyelination is a reversible phenomenon, and drugs such as 4-aminopyridine (4AP), which target K+ channels in demyelinated axons, show that conduction can be restored. Yet, accurately assessing and monitoring demyelination after SCI remains challenging because of the lack of suitable imaging methods. In this study, we introduce a novel approach using the PET tracer, 3-[18F]fluoro-4-aminopyridine ([18F]3F4AP), specifically targeting K+ channels in demyelinated axons for SCI imaging. Methods: Rats with incomplete contusion injuries were imaged with [18F]3F4AP PET up to 1 mo after injury, followed by further validation of PET imaging results with autoradiography and immunohistochemistry of postmortem spinal cord tissue. A proof-of-concept study in 2 human subjects with incomplete injuries of different severities and etiologies was also performed. Results: [18F]3F4AP PET of SCI rats revealed a more than 2-fold increase in tracer binding highly localized to the injured segment of the cord at 7 d after injury relative to baseline (SUV ratio = 2.49 ± 0.09 for 7 d after injury vs. 1.14 ± 0.10 for baseline), revealing [18F]3F4AP's exceptional sensitivity to injury and its ability to detect temporal changes. Autoradiography, histology, and immunohistochemistry confirmed [18F]3F4AP's targeting of demyelinated axons. In humans, [18F]3F4AP differentiated between a severe and a largely recovered incomplete injury, indicating axonal loss and demyelination, respectively. Moreover, alterations in tracer delivery were evident on dynamic PET images, suggestive of differences in spinal cord blood flow between the injuries. Conclusion: [18F]3F4AP demonstrates efficacy in detecting incomplete SCI in both animal models and humans. The potential for monitoring post-SCI demyelination changes and response to therapy underscores the utility of [18F]3F4AP in advancing our understanding and management of SCI.

Quantitative Accuracy Assessment of the NeuroEXPLORER for Diverse Imaging Applications: Moving Beyond Standard Evaluations

Quantitative molecular imaging with PET can offer insights into physiologic and pathologic processes and is widely used for studying brain disorders. The NeuroEXPLORER is a recently developed dedicated brain PET system offering high spatial resolution and high sensitivity with an extended axial length. This study evaluated the quantitative precision and accuracy of the NeuroEXPLORER with phantom and human data for a variety of imaging conditions that are relevant to dynamic neuroimaging studies. Methods: Thirty-minute scans of an image quality (IQ) phantom and a 3-dimensional Hoffman brain phantom filled with [18F]FDG were performed over 13 h, covering phantom activities of 1.3-177 MBq. Furthermore, a uniform cylindric phantom filled with 558 MBq of 11C was scanned for 4 h. Quantitative accuracy was assessed using the contrast recovery coefficient (CRC), background variability, and background bias in the IQ phantom, the recovery coefficients (RCs) in the Hoffman phantom, and the bias in the uniform phantom. Results were compared at delayed time points, with different reconstruction parameters and frame lengths down to 1 s. Moreover, randomly subsampled frames of 2 imaging time points (0-2 min and 60-90 min) from a dynamic scan of a healthy volunteer with a 177-MBq injected dose of (R)-4-(3-fluoro-5-(fluoro-18F)phenyl)-1-((3-methylpyridin-4-yl)methyl)pyrrolidin-2-one ([18F]SynVesT-1) were used to assess quantification of brain uptake and image-derived input function extraction. Results: Negligible effects were observed on CRC and background bias with 3-177 MBq in the IQ phantom, and bias was less than 5% with 1-558 MBq in the uniform phantom. RC variations were within ±1% with 2-169 MBq in the Hoffman phantom, showcasing the system's high spatial resolution and high sensitivity. Short-frame reconstructions of the 60- to 90-min healthy-volunteer scan showed a ±1% mean difference in quantification of brain uptake for frame lengths down to 30 s and demonstrated the feasibility of measuring image-derived input function with mean absolute differences below 10% for frame lengths down to 1 s. Conclusion: The NeuroEXPLORER, with its high detection sensitivity, maintains high precision and accuracy across a wide range of imaging conditions beyond those evaluated in standard performance tests. These results demonstrate its potential for quantitative neuroimaging applications.

PET image reconstruction using weighted nuclear norm maximization and deep learning prior

The ill-posed Positron emission tomography (PET) reconstruction problem usually results in limited resolution and significant noise. Recently, deep neural networks have been incorporated into PET iterative reconstruction framework to improve the image quality. In this paper, we propose a new neural network-based iterative reconstruction method by using weighted nuclear norm (WNN) maximization, which aims to recover the image details in the reconstruction process. The novelty of our method is the application of WNN maximization rather than WNN minimization in PET image reconstruction. Meanwhile, a neural network is used to control the noise originated from WNN maximization. Our method is evaluated on simulated and clinical datasets. The simulation results show that the proposed approach outperforms state-of-the-art neural network-based iterative methods by achieving the best contrast/noise tradeoff with a remarkable contrast improvement on the lesion contrast recovery. The study on clinical datasets also demonstrates that our method can recover lesions of different sizes while suppressing noise in various low-dose PET image reconstruction tasks. Our code is available athttps://github.com/Kuangxd/PETReconstruction.

Visualizing scRNA-Seq data at population scale with GloScope

Increasingly, scRNA-Seq studies explore cell populations across different samples and the effect of sample heterogeneity on organism's phenotype. However, relatively few bioinformatic methods have been developed which adequately address the variation between samples for such population-level analyses. We propose a framework for representing the entire single-cell profile of a sample, which we call a GloScope representation. We implement GloScope on scRNA-Seq datasets from study designs ranging from 12 to over 300 samples and demonstrate how GloScope allows researchers to perform essential bioinformatic tasks at the sample-level, in particular visualization and quality control assessment.

Radiation image reconstruction and uncertainty quantification using a Gaussian process prior

We propose a complete framework for Bayesian image reconstruction and uncertainty quantification based on a Gaussian process prior (GPP) to overcome limitations of maximum likelihood expectation maximization (ML-EM) image reconstruction algorithm. The prior distribution is constructed with a zero-mean Gaussian process (GP) with a choice of a covariance function, and a link function is used to map the Gaussian process to an image. Unlike many other maximum a posteriori approaches, our method offers highly interpretable hyperparamters that are selected automatically with the empirical Bayes method. Furthermore, the GP covariance function can be modified to incorporate a priori structural priors, enabling multi-modality imaging or contextual data fusion. Lastly, we illustrate that our approach lends itself to Bayesian uncertainty quantification techniques, such as the preconditioned Crank-Nicolson method and the Laplace approximation. The proposed framework is general and can be employed in most radiation image reconstruction problems, and we demonstrate it with simulated free-moving single detector radiation source imaging scenarios. We compare the reconstruction results from GPP and ML-EM, and show that the proposed method can significantly improve the image quality over ML-EM, all the while providing greater understanding of the source distribution via the uncertainty quantification capability. Furthermore, significant improvement of the image quality by incorporating a structural prior is illustrated.

Calculation of intravoxel incoherent motion parameter maps using a kernelized total difference-based method

Quantitative analysis of diffusion-weighted magnetic resonance imaging (DW-MRI) has been explored for many clinical applications since its development. In particular, the intravoxel incoherent motion (IVIM) model for DW-MRI has been commonly utilized in various organs. However, because of the presence of excessive noise, the IVIM parameter maps obtained from pixel-wise fitting are often unreliable. In this study, we propose a kernelized total difference-based curve-fitting method to estimate the IVIM parameters. Simulated DW-MRI data at five signal-to-noise ratios (i.e., 10, 20, 30, 50, and 100) and real abdominal DW-MRI data acquired on a 1.5-T MRI scanner with nine b-values (i.e., 0, 10, 25, 50, 100, 200, 300, 400, and 500 s/mm2) and six diffusion-encoding gradient directions were used to evaluate the performance of the proposed method. The results were compared with those obtained by three existing methods: trust-region reflective (TRR) algorithm, Bayesian probability (BP), and deep neural network (DNN). Our simulation results showed that the proposed method outperformed the other three comparing methods in terms of root-mean-square error. Moreover, the proposed method could preserve small details in the estimated IVIM parameter maps. The experimental results showed that, compared with the TRR method, the proposed method as well as the BP (and DNN) method could reduce the overestimation of the pseudodiffusion coefficient and improve the quality of IVIM parameter maps. For all studied abdominal organs except the pancreas, both the proposed method and the BP method could provide IVIM parameter estimates close to the reference values; the former had higher precision. The kernelized total difference-based curve-fitting method has the potential to improve the reliability of IVIM parametric imaging.

Systematic Review on Learning-based Spectral CT

Spectral computed tomography (CT) has recently emerged as an advanced version of medical CT and significantly improves conventional (single-energy) CT. Spectral CT has two main forms: dual-energy computed tomography (DECT) and photon-counting computed tomography (PCCT), which offer image improvement, material decomposition, and feature quantification relative to conventional CT. However, the inherent challenges of spectral CT, evidenced by data and image artifacts, remain a bottleneck for clinical applications. To address these problems, machine learning techniques have been widely applied to spectral CT. In this review, we present the state-of-the-art data-driven techniques for spectral CT.

Total-Body Dynamic Imaging and Kinetic Modeling of [18F]F-AraG in Healthy Individuals and a Non-Small Cell Lung Cancer Patient Undergoing Anti-PD-1 Immunotherapy

Immunotherapies, especially checkpoint inhibitors such as anti-programmed cell death protein 1 (anti-PD-1) antibodies, have transformed cancer treatment by enhancing the immune system's capability to target and kill cancer cells. However, predicting immunotherapy response remains challenging. 18F-arabinosyl guanine ([18F]F-AraG) is a molecular imaging tracer targeting activated T cells, which may facilitate therapy response assessment by noninvasive quantification of immune cell activity within the tumor microenvironment and elsewhere in the body. The aim of this study was to obtain preliminary data on total-body pharmacokinetics of [18F]F-AraG as a potential quantitative biomarker for immune response evaluation. Methods: The study consisted of 90-min total-body dynamic scans of 4 healthy subjects and 1 non-small cell lung cancer patient who was scanned before and after anti-PD-1 immunotherapy. Compartmental modeling with Akaike information criterion model selection was used to analyze tracer kinetics in various organs. Additionally, 7 subregions of the primary lung tumor and 4 mediastinal lymph nodes were analyzed. Practical identifiability analysis was performed to assess the reliability of kinetic parameter estimation. Correlations of the SUVmean, the tissue-to-blood SUV ratio (SUVR), and the Logan plot slope (K Logan) with the total volume of distribution (V T) were calculated to identify potential surrogates for kinetic modeling. Results: Strong correlations were observed between K Logan and SUVR with V T, suggesting that they can be used as promising surrogates for V T, especially in organs with a low blood-volume fraction. Moreover, practical identifiability analysis suggested that dynamic [18F]F-AraG PET scans could potentially be shortened to 60 min, while maintaining quantification accuracy for all organs of interest. The study suggests that although [18F]F-AraG SUV images can provide insights on immune cell distribution, kinetic modeling or graphical analysis methods may be required for accurate quantification of immune response after therapy. Although SUVmean showed variable changes in different subregions of the tumor after therapy, the SUVR, K Logan, and V T showed consistent increasing trends in all analyzed subregions of the tumor with high practical identifiability. Conclusion: Our findings highlight the promise of [18F]F-AraG dynamic imaging as a noninvasive biomarker for quantifying the immune response to immunotherapy in cancer patients. Promising total-body kinetic modeling results also suggest potentially wider applications of the tracer in investigating the role of T cells in the immunopathogenesis of diseases.

Quantitative Total-Body Imaging of Blood Flow with High Temporal Resolution Early Dynamic 18F-Fluorodeoxyglucose PET Kinetic Modeling

Quantitative total-body PET imaging of blood flow can be performed with freely diffusible flow radiotracers such as 15O-water and 11C-butanol, but their short half-lives necessitate close access to a cyclotron. Past efforts to measure blood flow with the widely available radiotracer 18F-fluorodeoxyglucose (FDG) were limited to tissues with high 18F-FDG extraction fraction. In this study, we developed an early-dynamic 18F-FDG PET method with high temporal resolution kinetic modeling to assess total-body blood flow based on deriving the vascular transit time of 18F-FDG and conducted a pilot comparison study against a 11C-butanol reference.

Methods

The first two minutes of dynamic PET scans were reconstructed at high temporal resolution (60×1 s, 30×2 s) to resolve the rapid passage of the radiotracer through blood vessels. In contrast to existing methods that use blood-to-tissue transport rate (K 1 ) as a surrogate of blood flow, our method directly estimates blood flow using a distributed kinetic model (adiabatic approximation to the tissue homogeneity model; AATH). To validate our 18F-FDG measurements of blood flow against a flow radiotracer, we analyzed total-body dynamic PET images of six human participants scanned with both 18F-FDG and 11C-butanol. An additional thirty-four total-body dynamic 18F-FDG PET scans of healthy participants were analyzed for comparison against literature blood flow ranges. Regional blood flow was estimated across the body and total-body parametric imaging of blood flow was conducted for visual assessment. AATH and standard compartment model fitting was compared by the Akaike Information Criterion at different temporal resolutions.

Results

18F-FDG blood flow was in quantitative agreement with flow measured from 11C-butanol across same-subject regional measurements (Pearson R=0.955, p<0.001; linear regression y=0.973x-0.012), which was visually corroborated by total-body blood flow parametric imaging. Our method resolved a wide range of blood flow values across the body in broad agreement with literature ranges (e.g., healthy cohort average: 0.51±0.12 ml/min/cm3 in the cerebral cortex and 2.03±0.64 ml/min/cm3 in the lungs, respectively). High temporal resolution (1 to 2 s) was critical to enabling AATH modeling over standard compartment modeling.

Conclusions

Total-body blood flow imaging was feasible using early-dynamic 18F-FDG PET with high-temporal resolution kinetic modeling. Combined with standard 18F-FDG PET methods, this method may enable efficient single-tracer flow-metabolism imaging, with numerous research and clinical applications in oncology, cardiovascular disease, pain medicine, and neuroscience.

Visualizing scRNA-Seq Data at Population Scale with GloScope

Increasingly, scRNA-Seq studies explore cell populations across different samples and the effect of sample heterogeneity on organism's phenotype. However, relatively few bioinformatic methods have been developed which adequately address the variation between samples for such population-level analyses. We propose a framework for representing the entire single-cell profile of a sample, which we call a GloScope representation. We implement GloScope on scRNA-Seq datasets from study designs ranging from 12 to over 300 samples and demonstrate how GloScope allows researchers to perform essential bioinformatic tasks at the sample-level, in particular visualization and quality control assessment.

Quantitative PET imaging and modeling of molecular blood-brain barrier permeability

Blood-brain barrier (BBB) disruption is involved in the pathogenesis and progression of many neurological and systemic diseases. Non-invasive assessment of BBB permeability in humans has mainly been performed with dynamic contrast-enhanced magnetic resonance imaging, evaluating the BBB as a structural barrier. Here, we developed a novel non-invasive positron emission tomography (PET) method in humans to measure the BBB permeability of molecular radiotracers that cross the BBB through different transport mechanisms. Our method uses high-temporal resolution dynamic imaging and kinetic modeling to jointly estimate cerebral blood flow and tracer-specific BBB transport rate from a single dynamic PET scan and measure the molecular permeability-surface area (PS) product of the radiotracer. We show our method can resolve BBB PS across three PET radiotracers with greatly differing permeabilities, measure reductions in BBB PS of 18F-fluorodeoxyglucose (FDG) in healthy aging, and demonstrate a possible brain-body association between decreased FDG BBB PS in patients with metabolic dysfunction-associated steatotic liver inflammation. Our method opens new directions to efficiently study the molecular permeability of the human BBB in vivo using the large catalogue of available molecular PET tracers.

Learning Disentangled Representation for Multidimensional MR Image Reconstruction

We proposed a new way to represent and reconstruct multidimensional MR images. Specifically, a representation capable of disentangling different types of features in high-dimensional images was learned via training an autoencoder with separated sets of latent spaces for image style transfer, e.g., contrast or geometry transfer. A latent diffusion model was introduced to capture the distributions of the disentangled latents for constrained reconstruction. A new formulation was developed to integrate the pre-learned representation with other complementary constraints for reconstruction from sparse data. We demonstrated the ability of our model to disentangle contrast and geometry features in multicontrast MR images, and its effectiveness in accelerated T1 and T2 mapping.

SPLIT: Statistical Positronium Lifetime Image Reconstruction via Time-Thresholding

Positron emission tomography (PET) is a widely utilized medical imaging modality that uses positron-emitting radiotracers to visualize biochemical processes in a living body. The spatiotemporal distribution of a radiotracer is estimated by detecting the coincidence photon pairs generated through positron annihilations. In human tissue, about 40% of the positrons form positroniums prior to the annihilation. The lifetime of these positroniums is influenced by the microenvironment in the tissue and could provide valuable information for better understanding of disease progression and treatment response. Currently, there are few methods available for reconstructing high-resolution lifetime images in practical applications. This paper presents an efficient statistical image reconstruction method for positronium lifetime imaging (PLI). We also analyze the random triple-coincidence events in PLI and propose a correction method for random events, which is essential for real applications. Both simulation and experimental studies demonstrate that the proposed method can produce lifetime images with high numerical accuracy, low variance, and resolution comparable to that of the activity images generated by a PET scanner with currently available time-of-flight resolution.

High-Dimensional MR Reconstruction Integrating Subspace and Adaptive Generative Models

OBJECTIVE

To develop a new method that integrates subspace and generative image models for high-dimensional MR image reconstruction.

METHODS

We proposed a formulation that synergizes a low-dimensional subspace model of high-dimensional images, an adaptive generative image prior serving as spatial constraints on the sequence of "contrast-weighted" images or spatial coefficients of the subspace model, and a conventional sparsity regularization. A special pretraining plus subject-specific network adaptation strategy was proposed to construct an accurate generative-network-based representation for images with varying contrasts. An iterative algorithm was introduced to jointly update the subspace coefficients and the multi-resolution latent space of the generative image model that leveraged an recently proposed intermediate layer optimization technique for network inversion.

RESULTS

We evaluated the utility of the proposed method for two high-dimensional imaging applications: accelerated MR parameter mapping and high-resolution MR spectroscopic imaging. Improved performance over state-of-the-art subspace-based methods was demonstrated in both cases.

CONCLUSION

The proposed method provided a new way to address high-dimensional MR image reconstruction problems by incorporating an adaptive generative model as a data-driven spatial prior for constraining subspace reconstruction.

SIGNIFICANCE

Our work demonstrated the potential of integrating data-driven and adaptive generative priors with canonical low-dimensional modeling for high-dimensional imaging problems.

Optimization-derived blood input function using a kernel method and its evaluation with total-body PET for brain parametric imaging

Dynamic PET allows quantification of physiological parameters through tracer kinetic modeling. For dynamic imaging of brain or head and neck cancer on conventional PET scanners with a short axial field of view, the image-derived input function (ID-IF) from intracranial blood vessels such as the carotid artery (CA) suffers from severe partial volume effects. Alternatively, optimization-derived input function (OD-IF) by the simultaneous estimation (SIME) method does not rely on an ID-IF but derives the input function directly from the data. However, the optimization problem is often highly ill-posed. We proposed a new method that combines the ideas of OD-IF and ID-IF together through a kernel framework. While evaluation of such a method is challenging in human subjects, we used the uEXPLORER total-body PET system that covers major blood pools to provide a reference for validation.

METHODS

The conventional SIME approach estimates an input function using a joint estimation together with kinetic parameters by fitting time activity curves from multiple regions of interests (ROIs). The input function is commonly parameterized with a highly nonlinear model which is difficult to estimate. The proposed kernel SIME method exploits the CA ID-IF as a priori information via a kernel representation to stabilize the SIME approach. The unknown parameters are linear and thus easier to estimate. The proposed method was evaluated using 18F-fluorodeoxyglucose studies with both computer simulations and 20 human-subject scans acquired on the uEXPLORER scanner. The effect of the number of ROIs on kernel SIME was also explored.

RESULTS

The estimated OD-IF by kernel SIME showed a good match with the reference input function and provided more accurate estimation of kinetic parameters for both simulation and human-subject data. The kernel SIME led to the highest correlation coefficient (R = 0.97) and the lowest mean absolute error (MAE = 10.5 %) compared to using the CA ID-IF (R = 0.86, MAE = 108.2 %) and conventional SIME (R = 0.57, MAE = 78.7 %) in the human-subject evaluation. Adding more ROIs improved the overall performance of the kernel SIME method.

CONCLUSION

The proposed kernel SIME method shows promise to provide an accurate estimation of the blood input function and kinetic parameters for brain PET parametric imaging.

Anatomically Guided PET Image Reconstruction Using Conditional Weakly-Supervised Multi-Task Learning Integrating Self-Attention

To address the lack of high-quality training labels in positron emission tomography (PET) imaging, weakly-supervised reconstruction methods that generate network-based mappings between prior images and noisy targets have been developed. However, the learned model has an intrinsic variance proportional to the average variance of the target image. To suppress noise and improve the accuracy and generalizability of the learned model, we propose a conditional weakly-supervised multi-task learning (MTL) strategy, in which an auxiliary task is introduced serving as an anatomical regularizer for the PET reconstruction main task. In the proposed MTL approach, we devise a novel multi-channel self-attention (MCSA) module that helps learn an optimal combination of shared and task-specific features by capturing both local and global channel-spatial dependencies. The proposed reconstruction method was evaluated on NEMA phantom PET datasets acquired at different positions in a PET/CT scanner and 26 clinical whole-body PET datasets. The phantom results demonstrate that our method outperforms state-of-the-art learning-free and weakly-supervised approaches obtaining the best noise/contrast tradeoff with a significant noise reduction of approximately 50.0% relative to the maximum likelihood (ML) reconstruction. The patient study results demonstrate that our method achieves the largest noise reductions of 67.3% and 35.5% in the liver and lung, respectively, as well as consistently small biases in 8 tumors with various volumes and intensities. In addition, network visualization reveals that adding the auxiliary task introduces more anatomical information into PET reconstruction than adding only the anatomical loss, and the developed MCSA can abstract features and retain PET image details.

Imaging demyelinated axons after spinal cord injuries with PET tracer [ 18 F]3F4AP

Spinal cord injuries (SCI) often lead to lifelong disability. Among the various types of injuries, incomplete and discomplete injuries, where some axons remain intact, offer potential for recovery. However, demyelination of these spared axons can worsen disability. Demyelination is a reversible phenomenon, and drugs like 4-aminopyridine (4AP), which target K+ channels in demyelinated axons, show that conduction can be restored. Yet, accurately assessing and monitoring demyelination post-SCI remains challenging due to the lack of suitable imaging methods. In this study, we introduce a novel approach utilizing the positron emission tomography (PET) tracer, [ 18 F]3F4AP, specifically targeting K+ channels in demyelinated axons for SCI imaging. Rats with incomplete contusion injuries were imaged up to one month post-injury, revealing [ 18 F]3F4AP's exceptional sensitivity to injury and its ability to detect temporal changes. Further validation through autoradiography and immunohistochemistry confirmed [ 18 F]3F4AP's targeting of demyelinated axons. In a proof-of-concept study involving human subjects, [ 18 F]3F4AP differentiated between a severe and a largely recovered incomplete injury, indicating axonal loss and demyelination, respectively. Moreover, alterations in tracer delivery were evident on dynamic PET images, suggestive of differences in spinal cord blood flow between the injuries. In conclusion, [ 18 F]3F4AP demonstrates efficacy in detecting incomplete SCI in both animal models and humans. The potential for monitoring post-SCI demyelination changes and response to therapy underscores the utility of [ 18 F]3F4AP in advancing our understanding and management of spinal cord injuries.

Efficient image reconstruction for a small animal PET system with dual-layer-offset detector design

BACKGROUND

A compact PET/SPECT/CT system Inliview-3000B has been developed to provide multi-modality information on small animals for biomedical research. Its PET subsystem employed a dual-layer-offset detector design for depth-of-interaction capability and higher detection efficiency, but the irregular design caused some difficulties in calculating the normalization factors and the sensitivity map. Besides, the relatively larger (2 mm) crystal cross-section size also posed a challenge to high-resolution image reconstruction.

PURPOSE

We present an efficient image reconstruction method to achieve high imaging performance for the PET subsystem of Inliview-3000B.

METHODS

List mode reconstruction with efficient system modeling was used for the PET imaging. We adopt an on-the-fly multi-ray tracing method with random crystal sampling to model the solid angle, crystal penetration and object attenuation effect, and modify the system response model during each iteration to improve the reconstruction performance and computational efficiency. We estimate crystal efficiency with a novel iterative approach that combines measured cylinder phantom data with simulated line-of-response (LOR)-based factors for normalization correction before reconstruction. Since it is necessary to calculate normalization factors and the sensitivity map, we stack the two crystal layers together and extend the conventional data organization method here to index all useful LORs. Simulations and experiments were performed to demonstrate the feasibility and advantage of the proposed method.

RESULTS

Simulation results showed that the iterative algorithm for crystal efficiency estimation could achieve good accuracy. NEMA image quality phantom studies have demonstrated the superiority of random sampling, which is able to achieve good imaging performance with much less computation than traditional uniform sampling. In the spatial resolution evaluation based on the mini-Derenzo phantom, 1.1 mm hot rods could be identified with the proposed reconstruction method. Reconstruction of double mice and a rat showed good spatial resolution and a high signal-to-noise ratio, and organs with higher uptake could be recognized well.

CONCLUSION

The results validated the superiority of introducing randomness into reconstruction, and demonstrated its reliability for high-performance imaging. The Inliview-3000B PET subsystem with the proposed image reconstruction can provide rich and detailed information on small animals for preclinical research.

An iterative reconstruction algorithm for unsupervised PET image

Objective.In recent years, convolutional neural networks (CNNs) have shown great potential in positron emission tomography (PET) image reconstruction. However, most of them rely on many low-quality and high-quality reference PET image pairs for training, which are not always feasible in clinical practice. On the other hand, many works improve the quality of PET image reconstruction by adding explicit regularization or optimizing the network structure, which may lead to complex optimization problems.Approach.In this paper, we develop a novel iterative reconstruction algorithm by integrating the deep image prior (DIP) framework, which only needs the prior information (e.g. MRI) and sinogram data of patients. To be specific, we construct the objective function as a constrained optimization problem and utilize the existing PET image reconstruction packages to streamline calculations. Moreover, to further improve both the reconstruction quality and speed, we introduce the Nesterov's acceleration part and the restart mechanism in each iteration.Main results.2D experiments on PET data sets based on computer simulations and real patients demonstrate that our proposed algorithm can outperform existing MLEM-GF, KEM and DIPRecon methods.Significance.Unlike traditional CNN methods, the proposed algorithm does not rely on large data sets, but only leverages inter-patient information. Furthermore, we enhance reconstruction performance by optimizing the iterative algorithm. Notably, the proposed method does not require much modification of the basic algorithm, allowing for easy integration into standard implementations.

Systematic Review on Learning-based Spectral CT

Spectral computed tomography (CT) has recently emerged as an advanced version of medical CT and significantly improves conventional (single-energy) CT. Spectral CT has two main forms: dual-energy computed tomography (DECT) and photon-counting computed tomography (PCCT), which offer image improvement, material decomposition, and feature quantification relative to conventional CT. However, the inherent challenges of spectral CT, evidenced by data and image artifacts, remain a bottleneck for clinical applications. To address these problems, machine learning techniques have been widely applied to spectral CT. In this review, we present the state-of-the-art data-driven techniques for spectral CT.

Recent advances in combined Positron Emission Tomography and Magnetic Resonance Imaging

Hybrid imaging modalities combine two or more medical imaging techniques offering exciting new possibilities to image the structure, function and biochemistry of the human body in far greater detail than has previously been possible to improve patient diagnosis. In this context, simultaneous Positron Emission Tomography and Magnetic Resonance (PET/MR) imaging offers great complementary information, but it also poses challenges from the point of view of hardware and software compatibility. The PET signal may interfere with the MR magnetic field and vice-versa, posing several challenges and constrains in the PET instrumentation for PET/MR systems. Additionally, anatomical maps are needed to properly apply attenuation and scatter corrections to the resulting reconstructed PET images, as well motion estimates to minimize the effects of movement throughout the acquisition. In this review, we summarize the instrumentation implemented in modern PET scanners to overcome these limitations, describing the historical development of hybrid PET/MR scanners. We pay special attention to the methods used in PET to achieve attenuation, scatter and motion correction when it is combined with MR, and how both imaging modalities may be combined in PET image reconstruction algorithms.

Deep Generalized Learning Model for PET Image Reconstruction

Low-count positron emission tomography (PET) imaging is challenging because of the ill-posedness of this inverse problem. Previous studies have demonstrated that deep learning (DL) holds promise for achieving improved low-count PET image quality. However, almost all data-driven DL methods suffer from fine structure degradation and blurring effects after denoising. Incorporating DL into the traditional iterative optimization model can effectively improve its image quality and recover fine structures, but little research has considered the full relaxation of the model, resulting in the performance of this hybrid model not being sufficiently exploited. In this paper, we propose a learning framework that deeply integrates DL and an alternating direction of multipliers method (ADMM)-based iterative optimization model. The innovative feature of this method is that we break the inherent forms of the fidelity operators and use neural networks to process them. The regularization term is deeply generalized. The proposed method is evaluated on simulated data and real data. Both the qualitative and quantitative results show that our proposed neural network method can outperform partial operator expansion-based neural network methods, neural network denoising methods and traditional methods.

Subspace Model-Assisted Deep Learning for Improved Image Reconstruction

Image reconstruction from limited and/or sparse data is known to be an ill-posed problem and a priori information/constraints have played an important role in solving the problem. Early constrained image reconstruction methods utilize image priors based on general image properties such as sparsity, low-rank structures, spatial support bound, etc. Recent deep learning-based reconstruction methods promise to produce even higher quality reconstructions by utilizing more specific image priors learned from training data. However, learning high-dimensional image priors requires huge amounts of training data that are currently not available in medical imaging applications. As a result, deep learning-based reconstructions often suffer from two known practical issues: a) sensitivity to data perturbations (e.g., changes in data sampling scheme), and b) limited generalization capability (e.g., biased reconstruction of lesions). This paper proposes a new method to address these issues. The proposed method synergistically integrates model-based and data-driven learning in three key components. The first component uses the linear vector space framework to capture global dependence of image features; the second exploits a deep network to learn the mapping from a linear vector space to a nonlinear manifold; the third is an unrolling-based deep network that captures local residual features with the aid of a sparsity model. The proposed method has been evaluated with magnetic resonance imaging data, demonstrating improved reconstruction in the presence of data perturbation and/or novel image features. The method may enhance the practical utility of deep learning-based image reconstruction.

Total-body Dynamic Imaging and Kinetic Modeling of 18F-AraG in Healthy Individuals and a Non-Small Cell Lung Cancer Patient Undergoing Anti-PD-1 Immunotherapy

Immunotherapies, especially the checkpoint inhibitors such as anti-PD-1 antibodies, have transformed cancer treatment by enhancing immune system's capability to target and kill cancer cells. However, predicting immunotherapy response remains challenging. 18F-AraG is a molecular imaging tracer targeting activated T cells, which may facilitate therapy response assessment by non-invasive quantification of immune cell activity within tumor microenvironment and elsewhere in the body. The aim of this study was to obtain preliminary data on total-body pharmacokinetics of 18F-AraG, as a potential quantitative biomarker for immune response evaluation.

Methods

The study consisted of 90-min total-body dynamic scans of four healthy subjects and one non-small cell lung cancer (NSCLC) patient, scanned before and after anti-PD-1 immunotherapy. Compartmental modeling with Akaike information criterion model selection were employed to analyze tracer kinetics in various organs. Additionally, seven sub-regions of the primary lung tumor and four mediastinal lymph nodes were analyzed. Practical identifiability analysis was performed to assess reliability of kinetic parameter estimation. Correlations of SUVmean, SUVR (tissue-to-blood ratio), and Logan plot slope K L o g a n with total volume-of-distribution V T were calculated to identify potential surrogates for kinetic modeling.

Results

Strong correlations were observed between K L o g a n and SUVR values with V T , suggesting that they can be used as promising surrogates for V T , especially in organs with low blood-volume fraction. Moreover, the practical identifiability analysis suggests that the dynamic 18F-AraG PET scans could potentially be shortened to 60 minutes, while maintaining quantification accuracy for all organs-of-interest. The study suggests that although 18F-AraG SUV images can provide insights on immune cell distribution, kinetic modeling or graphical analysis methods may be required for accurate quantification of immune response post-therapy. While SUVmean showed variable changes in different sub-regions of the tumor post-therapy, the SUVR, K L o g a n , and V T showed consistent increasing trends in all analyzed sub-regions of the tumor with high practical identifiability.

Conclusion

Our findings highlight the promise of 18F-AraG dynamic imaging as a non-invasive biomarker for quantifying the immune response to immunotherapy in cancer patients. The promising total-body kinetic modeling results also suggest potentially wider applications of the tracer in investigating the role of T cells in the immunopathogenesis of diseases.

AI for PET image reconstruction

Image reconstruction for positron emission tomography (PET) has been developed over many decades, with advances coming from improved modelling of the data statistics and improved modelling of the imaging physics. However, high noise and limited spatial resolution have remained issues in PET imaging, and state-of-the-art PET reconstruction has started to exploit other medical imaging modalities (such as MRI) to assist in noise reduction and enhancement of PET's spatial resolution. Nonetheless, there is an ongoing drive towards not only improving image quality, but also reducing the injected radiation dose and reducing scanning times. While the arrival of new PET scanners (such as total body PET) is helping, there is always a need to improve reconstructed image quality due to the time and count limited imaging conditions. Artificial intelligence (AI) methods are now at the frontier of research for PET image reconstruction. While AI can learn the imaging physics as well as the noise in the data (when given sufficient examples), one of the most common uses of AI arises from exploiting databases of high-quality reference examples, to provide advanced noise compensation and resolution recovery. There are three main AI reconstruction approaches: (i) direct data-driven AI methods which rely on supervised learning from reference data, (ii) iterative (unrolled) methods which combine our physics and statistical models with AI learning from data, and (iii) methods which exploit AI with our known models, but crucially can offer benefits even in the absence of any example training data whatsoever. This article reviews these methods, considering opportunities and challenges of AI for PET reconstruction.

A depth-of-interaction rebinning method based on both geometric and activity weights

BACKGROUND AND OBJECTIVE

For positron emission tomography (PET) scanners with depth-of-interaction (DOI) measurement, the DOI rebinning method that utilizes DOI information to process the projection data is critical to image quality. Current DOI rebinning methods map coincidence events onto the rebinned sinogram based on the correlation of lines of response (LOR). This study aims to incorporate prior radioactivity distribution of the imaging object into DOI rebinning to obtain better image quality.

METHODS

A DOI rebinning method based on both geometric and activity weights was proposed to assign coincidence events to the rebinned sinogram defined by a virtual ring. The geometric weights, representing the correlation between LORs, were calculated based on the areas of intersection. The activity weights, reflecting the activity distribution of the imaging object, were derived from the previous reconstructed image.

RESULTS

Monte Carlo simulation data from four phantoms, including the image quality phantom, Derenzo phantom, and two rat-like ROBY phantoms, was used to evaluate the proposed method. The recovery coefficient (RC), contrast recovery coefficient (CRC), structural similarity index measure (SSIM), and peak signal-to-noise ratio (PSNR) were used as image quality metrics. Compared to other DOI rebinning methods, the proposed method achieved the highest RC (maximum improvement of 32%) and CRC at the same noise level and was also optimal in terms of the SSIM and PSNR. Meanwhile, incorporating the prior activity distribution into DOI rebinning also improved the image reconstruction speed.

CONCLUSIONS

This work developed a new DOI rebinning method combining the correlation of LORs with the prior activity distribution, achieving relatively optimal image quality and reconstruction speed. Furthermore, it still needs to be evaluated on the actual equipment.

Fully 3D implementation of the end-to-end deep image prior-based PET image reconstruction using block iterative algorithm

Objective. Deep image prior (DIP) has recently attracted attention owing to its unsupervised positron emission tomography (PET) image reconstruction method, which does not require any prior training dataset. In this paper, we present the first attempt to implement an end-to-end DIP-based fully 3D PET image reconstruction method that incorporates a forward-projection model into a loss function.Approach. A practical implementation of a fully 3D PET image reconstruction could not be performed at present because of a graphics processing unit memory limitation. Consequently, we modify the DIP optimization to a block iteration and sequential learning of an ordered sequence of block sinograms. Furthermore, the relative difference penalty (RDP) term is added to the loss function to enhance the quantitative accuracy of the PET image.Main results. We evaluated our proposed method using Monte Carlo simulation with [18F]FDG PET data of a human brain and a preclinical study on monkey-brain [18F]FDG PET data. The proposed method was compared with the maximum-likelihood expectation maximization (EM), maximuma posterioriEM with RDP, and hybrid DIP-based PET reconstruction methods. The simulation results showed that, compared with other algorithms, the proposed method improved the PET image quality by reducing statistical noise and better preserved the contrast of brain structures and inserted tumors. In the preclinical experiment, finer structures and better contrast recovery were obtained with the proposed method.Significance.The results indicated that the proposed method could produce high-quality images without a prior training dataset. Thus, the proposed method could be a key enabling technology for the straightforward and practical implementation of end-to-end DIP-based fully 3D PET image reconstruction.

Triple modality image reconstruction of PET data using SPECT, PET, CT information increases lesion uptake in images of patients treated with radioembolization with 90 Y micro-spheres

PURPOSE

Nuclear medicine imaging modalities like computed tomography (CT), single photon emission CT (SPECT) and positron emission tomography (PET) are employed in the field of theranostics to estimate and plan the dose delivered to tumors and the surrounding tissues and to monitor the effect of the therapy. However, therapeutic radionuclides often provide poor images, which translate to inaccurate treatment planning and inadequate monitoring images. Multimodality information can be exploited in the reconstruction to enhance image quality. Triple modality PET/SPECT/CT scanners are particularly useful in this context due to the easier registration process between images. In this study, we propose to include PET, SPECT and CT information in the reconstruction of PET data. The method is applied to Yttrium-90 (90 Y) data.

METHODS

Data from a NEMA phantom filled with90 Y were used for validation. PET, SPECT and CT data from 10 patients treated with Selective Internal Radiation Therapy (SIRT) were used. Different combinations of prior images using the Hybrid kernelized expectation maximization were investigated in terms of VOI activity and noise suppression.

RESULTS

Our results show that triple modality PET reconstruction provides significantly higher uptake when compared to the method used as standard in the hospital and OSEM. In particular, using CT-guided SPECT images, as guiding information in the PET reconstruction significantly increases uptake quantification on tumoral lesions.

CONCLUSION

This work proposes the first triple modality reconstruction method and demonstrates up to 69% lesion uptake increase over standard methods with SIRT90 Y patient data. Promising results are expected for other radionuclide combination used in theranostic applications using PET and SPECT.

Anatomically guided reconstruction improves lesion quantitation and detectability in bone SPECT/CT

Bone single-photon emission computed tomography (SPECT)/computed tomography (CT) imaging suffers from poor spatial resolution, but the image quality can be improved during SPECT reconstruction by using anatomical information derived from CT imaging. The purpose of this work was to compare two different anatomically guided SPECT reconstruction methods to ordered subsets expectation maximization (OSEM) which is the most commonly used reconstruction method in nuclear medicine. The comparison was done in terms of lesion quantitation and lesion detectability. Anatomically guided Bayesian reconstruction (AMAP) and kernelized ordered subset expectation maximization (KEM) algorithms were implemented and compared against OSEM. Artificial lesions with a wide range of lesion-to-background contrasts were added to normal bone SPECT/CT studies. The quantitative accuracy was assessed by the error in lesion standardized uptake values and lesion detectability by the area under the receiver operating characteristic curve generated by a non-prewhitening matched filter. AMAP and KEM provided significantly better quantitative accuracy than OSEM at all contrast levels. Accuracy was the highest when SPECT lesions were matched to a lesion on CT. Correspondingly, AMAP and KEM also had significantly better lesion detectability than OSEM at all contrast levels and reconstructions with matching CT lesions performed the best. Quantitative differences between AMAP and KEM algorithms were minor. Visually AMAP and KEM images looked similar. Anatomically guided reconstruction improves lesion quantitation and detectability markedly compared to OSEM. Differences between AMAP and KEM algorithms were small and thus probably clinically insignificant.

Neural KEM: A Kernel Method With Deep Coefficient Prior for PET Image Reconstruction

Image reconstruction of low-count positron emission tomography (PET) data is challenging. Kernel methods address the challenge by incorporating image prior information in the forward model of iterative PET image reconstruction. The kernelized expectation-maximization (KEM) algorithm has been developed and demonstrated to be effective and easy to implement. A common approach for a further improvement of the kernel method would be adding an explicit regularization, which however leads to a complex optimization problem. In this paper, we propose an implicit regularization for the kernel method by using a deep coefficient prior, which represents the kernel coefficient image in the PET forward model using a convolutional neural-network. To solve the maximum-likelihood neural network-based reconstruction problem, we apply the principle of optimization transfer to derive a neural KEM algorithm. Each iteration of the algorithm consists of two separate steps: a KEM step for image update from the projection data and a deep-learning step in the image domain for updating the kernel coefficient image using the neural network. This optimization algorithm is guaranteed to monotonically increase the data likelihood. The results from computer simulations and real patient data have demonstrated that the neural KEM can outperform existing KEM and deep image prior methods.

Dynamic PET images denoising using spectral graph wavelet transform

Positron emission tomography (PET) is a non-invasive molecular imaging method for quantitative observation of physiological and biochemical changes in living organisms. The quality of the reconstructed PET image is limited by many different physical degradation factors. Various denoising methods including Gaussian filtering (GF) and non-local mean (NLM) filtering have been proposed to improve the image quality. However, image denoising usually blurs edges, of which high frequency components are filtered as noises. On the other hand, it is well-known that edges in a PET image are important to detection and recognition of a lesion. Denoising while preserving the edges of PET images remains an important yet challenging problem in PET image processing. In this paper, we propose a novel denoising method with good edge-preserving performance based on spectral graph wavelet transform (SGWT) for dynamic PET images denoising. We firstly generate a composite image from the entire time series, then perform SGWT on the PET images, and finally reconstruct the low graph frequency content to get the denoised dynamic PET images. Experimental results on simulation and in vivo data show that the proposed approach significantly outperforms the GF, NLM and graph filtering methods. Compared with deep learning-based method, the proposed method has the similar denoising performance, but it does not need lots of training data and has low computational complexity.

HYPR4D kernel method on TOF PET data with validations including image-derived input function

Background

Positron emission tomography (PET) images are typically noisy especially in dynamic imaging where the PET data are divided into a number of short temporal frames often with a low number of counts. As a result, image features such as contrast and time–activity curves are highly variable. Noise reduction in PET is thus essential. Typical noise reduction methods tend to not preserve image features/patterns (e.g. contrast and size dependent) accurately. In this work, we report the first application of our HYPR4D kernel method on time-of-flight (TOF) PET data (i.e. PSF-HYPR4D-K-TOFOSEM). The proposed HYPR4D kernel method makes use of the mean 4D high frequency features and inconsistent noise patterns over OSEM subsets as well as the low noise property of the early reconstruction updates to achieve prior-free de-noising. The method was implemented and tested on the GE SIGNA PET/MR and was compared to the TOF reconstructions with PSF resolution modeling available on the system, namely PSF-TOFOSEM with and without standard post filter and PSF-TOFBSREM (TOF Q.Clear) with various beta values (regularization strengths).

Results

Results from experimental contrast phantom and human subject data with various PET tracers showed that the proposed method provides more robust and accurate image features compared to other regularization methods. The preservation of contrast for the PSF-HYPR4D-K-TOFOSEM was observed to be better and less dependent on the contrast and size of the target structures as compared to TOF Q.Clear and PSF-TOFOSEM with filter. At the same contrast level, PSF-HYPR4D-K-TOFOSEM achieved better 4D noise suppression than other methods (e.g. >2 times lower noise than TOF Q.Clear at the highest contrast). We also present a novel voxel search method to obtain an image-derived input function (IDIF) and demonstrate that the obtained IDIF is the most quantitative w.r.t. the measured blood samples when the acquired data are reconstructed with PSF-HYPR4D-K-TOFOSEM.

Conclusions

The overall results support superior performance of the PSF-HYPR4D-K-TOFOSEM for TOF PET data and demonstrate that the proposed method is likely suitable for all imaging tasks including the generation of IDIF without requiring any prior information as well as further improving the effective sensitivity of the imaging system.

Deep Kernel Representation for Image Reconstruction in PET

Image reconstruction for positron emission tomography (PET) is challenging because of the ill-conditioned tomographic problem and low counting statistics. Kernel methods address this challenge by using kernel representation to incorporate image prior information in the forward model of iterative PET image reconstruction. Existing kernel methods construct the kernels commonly using an empirical process, which may lead to unsatisfactory performance. In this paper, we describe the equivalence between the kernel representation and a trainable neural network model. A deep kernel method is then proposed by exploiting a deep neural network to enable automated learning of an improved kernel model and is directly applicable to single subjects in dynamic PET. The training process utilizes available image prior data to form a set of robust kernels in an optimized way rather than empirically. The results from computer simulations and a real patient dataset demonstrate that the proposed deep kernel method can outperform the existing kernel method and neural network method for dynamic PET image reconstruction.

Using domain knowledge for robust and generalizable deep learning-based CT-free PET attenuation and scatter correction

Despite the potential of deep learning (DL)-based methods in substituting CT-based PET attenuation and scatter correction for CT-free PET imaging, a critical bottleneck is their limited capability in handling large heterogeneity of tracers and scanners of PET imaging. This study employs a simple way to integrate domain knowledge in DL for CT-free PET imaging. In contrast to conventional direct DL methods, we simplify the complex problem by a domain decomposition so that the learning of anatomy-dependent attenuation correction can be achieved robustly in a low-frequency domain while the original anatomy-independent high-frequency texture can be preserved during the processing. Even with the training from one tracer on one scanner, the effectiveness and robustness of our proposed approach are confirmed in tests of various external imaging tracers on different scanners. The robust, generalizable, and transparent DL development may enhance the potential of clinical translation.

Deep learning-based methods have been proposed to substitute CT-based PET attenuation and scatter correction to achieve CT-free PET imaging. Here, the authors present a simple way to integrate domain knowledge in deep learning for CT-free PET imaging.

Total-Body PET Multiparametric Imaging of Cancer Using a Voxelwise Strategy of Compartmental Modeling

Quantitative dynamic PET with compartmental modeling has the potential to enable multiparametric imaging and more accurate quantification than static PET imaging. Conventional methods for parametric imaging commonly use a single kinetic model for all image voxels and neglect the heterogeneity of physiologic models, which can work well for single-organ parametric imaging but may significantly compromise total-body parametric imaging on a scanner with a long axial field of view. In this paper, we evaluate the necessity of voxelwise compartmental modeling strategies, including time delay correction (TDC) and model selection, for total-body multiparametric imaging. Methods: Ten subjects (5 patients with metastatic cancer and 5 healthy volunteers) were scanned on a total-body PET/CT system after injection of 370 MBq of 18F-FDG. Dynamic data were acquired for 60 min. Total-body parametric imaging was performed using 2 approaches. One was the conventional method that uses a single irreversible 2-tissue-compartment model with and without TDC. The second approach selects the best kinetic model from 3 candidate models for individual voxels. The differences between the 2 approaches were evaluated for parametric imaging of microkinetic parameters and the 18F-FDG net influx rate, KiResults: TDC had a nonnegligible effect on kinetic quantification of various organs and lesions. The effect was larger in lesions with a higher blood volume. Parametric imaging of Ki with the standard 2-tissue-compartment model introduced vascular-region artifacts, which were overcome by the voxelwise model selection strategy. Conclusion: The time delay and appropriate kinetic model vary in different organs and lesions. Modeling of the time delay of the blood input function and model selection improved total-body multiparametric imaging.

Efficient Delay Correction for Total-Body PET Kinetic Modeling Using Pulse Timing Methods

Quantitative kinetic modeling requires an input function. A noninvasive image-derived input function (IDIF) can be obtained from dynamic PET images. However, a robust IDIF location (e.g., aorta) may be far from a tissue of interest, particularly in total-body PET, introducing a time delay between the IDIF and the tissue. The standard practice of joint estimation (JE) of delay, along with model fitting, is computationally expensive. To improve the efficiency of delay correction for total-body PET parametric imaging, this study investigated the use of pulse timing methods to estimate and correct for delay. Methods: Simulation studies were performed with a range of delay values, frame lengths, and noise levels to test the tolerance of 2 pulse timing methods-leading edge (LE) and constant fraction discrimination and their thresholds. The methods were then applied to data from 21 subjects (14 healthy volunteers, 7 cancer patients) who underwent a 60-min dynamic total-body ¹⁸F-FDG PET acquisition. Region-of-interest kinetic analysis was performed and parametric images were generated to compare LE and JE methods of delay correction, as well as no delay correction. Results: Simulations demonstrated that a 10% LE threshold resulted in biases and SDs at tolerable levels for all noise levels tested, with 2-s frames. Pooled region-of-interest-based results (n = 154) showed strong agreement between LE (10% threshold) and JE methods in estimating delay (Pearson r = 0.96, P < 0.001) and the kinetic parameters v_b (r = 0.96, P < 0.001), K_i (r = 1.00, P < 0.001), and K₁ (r = 0.97, P < 0.001). When tissues with minimal delay were excluded from pooled analyses, there were reductions in v_b (69.4%) and K₁ (4.8%) when delay correction was not performed. Similar results were obtained for parametric images; additionally, lesion K_i contrast was improved overall with LE and JE delay correction compared with no delay correction and Patlak analysis. Conclusion: This study demonstrated the importance of delay correction in total-body PET. LE delay correction can be an efficient surrogate for JE, requiring a fraction of the computational time and allowing for rapid delay correction across more than 10⁶ voxels in total-body PET datasets.

Spatiotemporal Kernel Reconstruction for Linear Parametric Neurotransmitter PET Kinetic Modeling in Motion Correction Brain PET of Awake Rats

The linear parametric neurotransmitter positron emission tomography (lp-ntPET) kinetic model can be used to detect transient changes (activation) in endogenous neurotransmitter levels. Preclinical PET scans in awake animals can be performed to investigate neurotransmitter transient changes. Here we use the spatiotemporal kernel reconstruction (Kernel) for noise reduction in dynamic PET, and lp-ntPET kinetic modeling. Kernel is adapted for motion correction reconstruction, applied in awake rat PET scans. We performed 2D rat brain phantom simulation using the ntPET model at 3 different noise levels. Data was reconstructed with independent frame reconstruction (IFR), IFR with HYPR denoising, and Kernel, and lp-ntPET kinetic parameters (k 2a : efflux rate, γ: activation magnitude, t d : activation onset time, and t p : activation peak time) were calculated. Additionally, significant activation magnitude (γ) difference with respect to a region with no activation (rest) was calculated. Finally, [11C]raclopride experiments were performed in anesthetized and awake rats, injecting cold raclopride at 20 min after scan start to simulate endogenous neurotransmitter release. For simulated data at the regional level, IFR coefficient of variation (COV) of k 2a , γ, t d and t p was reduced with HYPR denoising, but Kernel showed the lowest COV (2 fold reduction compared with IFR). At the pixel level the same trend is observed for k 2a , γ, t d and t p COV, but reduction is larger with Kernel compared with IFR (10-14 fold). Bias in γ with respect with noise-free values was additionally reduced using Kernel (difference of 292, 72.4, and -6.92% for IFR, IFR+KYPR, and Kernel, respectively). Significant difference in activation between the rest and active region could be detected at a simulated activation of 160% for IFR and IFR+HYPR, and of 120% for Kernel. In rat experiments, lp-ntPET parameters have better confidence intervals using Kernel. In the γ, and t d parametric maps, the striatum structure can be identified with Kernel but not with IFR. Striatum voxel-wise γ, t d and t p values have lower variability using Kernel compared with IFR and IFR+HYPR. The spatiotemporal kernel reconstruction adapted for motion correction reconstruction allows to improve lp-ntPET kinetic modeling noise in awake rat studies, as well as detection of subtle neurotransmitter activations.

Hybrid kernelised expectation maximisation for Bremsstrahlung SPECT reconstruction in SIRT with 90Y micro-spheres

Background

Selective internal radiation therapy with Yttrium-90 microspheres is an effective therapy for liver cancer and liver metastases. Yttrium-90 is mainly a high-energy beta particle emitter. These beta particles emit Bremsstrahlung radiation during their interaction with tissue making post-therapy imaging of the radioactivity distribution feasible. Nevertheless, image quality and quantification is difficult due to the continuous energy spectrum which makes resolution modelling, attenuation and scatter estimation challenging and therefore the dosimetry quantification is inaccurate. As a consequence a reconstruction algorithm able to improve resolution could be beneficial.

Methods

In this study, the hybrid kernelised expectation maximisation (HKEM) is used to improve resolution and contrast and reduce noise, in addition a modified HKEM called frozen HKEM (FHKEM) is investigated to further reduce noise. The iterative part of the FHKEM kernel was frozen at the 72nd sub-iteration. When using ordered subsets algorithms the data is divided in smaller subsets and the smallest algorithm iterative step is called sub-iteration. A NEMA phantom with spherical inserts was used for the optimisation and validation of the algorithm, and data from 5 patients treated with Selective internal radiation therapy were used as proof of clinical relevance of the method.

Results

The results suggest a maximum improvement of 56% for region of interest mean recovery coefficient at fixed coefficient of variation and better identification of the hot volumes in the NEMA phantom. Similar improvements were achieved with patient data, showing 47% mean value improvement over the gold standard used in hospitals.

Conclusions

Such quantitative improvements could facilitate improved dosimetry calculations with SPECT when treating patients with Selective internal radiation therapy, as well as provide a more visible position of the cancerous lesions in the liver.

Deep learning-based image reconstruction and post-processing methods in positron emission tomography for low-dose imaging and resolution enhancement

Image processing plays a crucial role in maximising diagnostic quality of positron emission tomography (PET) images. Recently, deep learning methods developed across many fields have shown tremendous potential when applied to medical image enhancement, resulting in a rich and rapidly advancing literature surrounding this subject. This review encapsulates methods for integrating deep learning into PET image reconstruction and post-processing for low-dose imaging and resolution enhancement. A brief introduction to conventional image processing techniques in PET is firstly presented. We then review methods which integrate deep learning into the image reconstruction framework as either deep learning-based regularisation or as a fully data-driven mapping from measured signal to images. Deep learning-based post-processing methods for low-dose imaging, temporal resolution enhancement and spatial resolution enhancement are also reviewed. Finally, the challenges associated with applying deep learning to enhance PET images in the clinical setting are discussed and future research directions to address these challenges are presented.

Accelerating MR Parameter Mapping Using Nonlinear Compressive Manifold Learning and Regularized Pre-Imaging

In this study, we presented a novel method to reconstruct the MR parametric maps from highly undersampled k-space data. Specifically, we utilized a nonlinear model to sparsely represent the unknown MR parameter-weighted images in high-dimensional feature space. Each image at a specific time point is assumed to belong to a low-dimensional manifold which is learned from training images created based on the parametric model. The final reconstruction is carried out by venturing the sparse representation of the images in the feature space back to the input space, using the pre-imaging technique. Particularly, among an infinite number of solutions that satisfy the data consistency, the one that is closest to the manifold is selected as the desired solution. The underlying optimization problem is solved using kernel trick, sparse coding, and split Bregman iteration algorithm. In addition, both spatial and temporal regularizations were utilized to further improve the reconstruction quality. The proposed method was validated on both phantom and in vivo human brain T2 mapping data. Results showed the proposed method was superior to the conventional linear model-based reconstruction methods, in terms of artifact removal and quantitative estimate accuracy. The proposed method could be potentially beneficial for quantitative MR applications.

Simultaneous Denoising of Dynamic PET Images Based on Deep Image Prior

Parametric imaging obtained from kinetic modeling analysis of dynamic positron emission tomography (PET) data is a useful tool for quantifying tracer kinetics. However, pixel-wise time-activity curves have high noise levels which lead to poor quality of parametric images. To solve this limitation, we proposed a new image denoising method based on deep image prior (DIP). Like the original DIP method, the proposed DIP method is an unsupervised method, in which no training dataset is required. However, the difference is that our method can simultaneously denoise all dynamic PET images. Moreover, we propose a modified version of the DIP method called double DIP (DDIP), which has two DIP architectures. The additional DIP model is used to generate high-quality input data for the second DIP model. Computer simulations were performed to evaluate the performance of the proposed DIP-based methods. Our simulation results showed that the DDIP method outperformed the single DIP method. In addition, the DDIP method combined with data augmentation could generate PET parametric images with superior image quality compared to the spatiotemporal-based non-local means filtering and high constrained backprojection. Our preliminary results show that our proposed DDIP method is a novel and effective unsupervised method for simultaneously denoising dynamic PET images.

Direct Reconstruction of Linear Parametric Images From Dynamic PET Using Nonlocal Deep Image Prior

Direct reconstruction methods have been developed to estimate parametric images directly from the measured PET sinograms by combining the PET imaging model and tracer kinetics in an integrated framework. Due to limited counts received, signal-to-noise-ratio (SNR) and resolution of parametric images produced by direct reconstruction frameworks are still limited. Recently supervised deep learning methods have been successfully applied to medical imaging denoising/reconstruction when large number of high-quality training labels are available. For static PET imaging, high-quality training labels can be acquired by extending the scanning time. However, this is not feasible for dynamic PET imaging, where the scanning time is already long enough. In this work, we proposed an unsupervised deep learning framework for direct parametric reconstruction from dynamic PET, which was tested on the Patlak model and the relative equilibrium Logan model. The training objective function was based on the PET statistical model. The patient's anatomical prior image, which is readily available from PET/CT or PET/MR scans, was supplied as the network input to provide a manifold constraint, and also utilized to construct a kernel layer to perform non-local feature denoising. The linear kinetic model was embedded in the network structure as a 1 ×1 ×1 convolution layer. Evaluations based on dynamic datasets of ¹⁸F-FDG and ¹¹C-PiB tracers show that the proposed framework can outperform the traditional and the kernel method-based direct reconstruction methods.

Dynamic PET Imaging Using Dual Texture Features

Purpose: This study aims to explore the impact of adding texture features in dynamic positron emission tomography (PET) reconstruction of imaging results. Methods: We have improved a reconstruction method that combines radiological dual texture features. In this method, multiple short time frames are added to obtain composite frames, and the image reconstructed by composite frames is used as the prior image. We extract texture features from prior images by using the gray level-gradient cooccurrence matrix (GGCM) and gray-level run length matrix (GLRLM). The prior information contains the intensity of the prior image, the inverse difference moment of the GGCM and the long-run low gray-level emphasis of the GLRLM. Results: The computer simulation results show that, compared with the traditional maximum likelihood, the proposed method obtains a higher signal-to-noise ratio (SNR) in the image obtained by dynamic PET reconstruction. Compared with similar methods, the proposed algorithm has a better normalized mean squared error (NMSE) and contrast recovery coefficient (CRC) at the tumor in the reconstructed image. Simulation studies on clinical patient images show that this method is also more accurate for reconstructing high-uptake lesions. Conclusion: By adding texture features to dynamic PET reconstruction, the reconstructed images are more accurate at the tumor.

Kernel graph filtering-A new method for dynamic sinogram denoising

Low count PET (positron emission tomography) imaging is often desirable in clinical diagnosis and biomedical research, but its images are generally very noisy, due to the very weak signals in the sinograms used in image reconstruction. To address this issue, this paper presents a novel kernel graph filtering method for dynamic PET sinogram denoising. This method is derived from treating the dynamic sinograms as the signals on a graph, and learning the graph adaptively from the kernel principal components of the sinograms to construct a lowpass kernel graph spectrum filter. The kernel graph filter thus obtained is then used to filter the original sinogram time frames to obtain the denoised sinograms for PET image reconstruction. Extensive tests and comparisons on the simulated and real life in-vivo dynamic PET datasets show that the proposed method outperforms the existing methods in sinogram denoising and image enhancement of dynamic PET at all count levels, especially at low count, with a great potential in real life applications of dynamic PET imaging.

Dynamic PET image reconstruction incorporating a median nonlocal means kernel method

In dynamic positron emission tomography (PET) imaging, the reconstructed image of a single frame often exhibits high noise due to limited counting statistics of projection data. This study proposed a median nonlocal means (MNLM)-based kernel method for dynamic PET image reconstruction. The kernel matrix is derived from median nonlocal means of pre-reconstructed composite images. Then the PET image intensities in all voxels were modeled as a kernel matrix multiplied by coefficients and incorporated into the forward model of PET projection data. Then, the coefficients of each feature were estimated by the maximum likelihood method. Using simulated low-count dynamic data of Zubal head phantom, the quantitative performance of the proposed MNLM kernel method was investigated and compared with the maximum-likelihood method, conventional kernel method with and without median filter, and nonlocal means (NLM) kernel method. Simulation results showed that the MNLM kernel method achieved visual and quantitative accuracy improvements (in terms of the ensemble mean squared error, bias versus variance, and contrast versus noise performances). Especially for frame 2 with the lowest count level of a single frame, the MNLM kernel method achieves lower ensemble mean squared error (10.43%) than the NLM kernel method (13.68%), conventional kernel method with and without median filter (11.88% and 23.50%), and MLEM algorithm (24.77%). The study on real low-dose ¹⁸F-FDG rat data also showed that the MNLM kernel method outperformed other methods in visual and quantitative accuracy improvements (in terms of regional noise versus intensity mean performance).