Dynamic PET denoising with HYPR processing.
Ontology highlight
ABSTRACT:
SUBMITTER: Christian BT
PROVIDER: S-EPMC3250311 | biostudies-literature | 2010 Jul
REPOSITORIES: biostudies-literature
Publications
Dynamic PET denoising with HYPR processing.
Christian Bradley T BT Vandehey Nicholas T NT Floberg John M JM Mistretta Charles A CA
Journal of nuclear medicine : official publication, Society of Nuclear Medicine 20100616 7
HighlY constrained backPRojection (HYPR) is a promising image-processing strategy with widespread application in time-resolved MRI that is also well suited for PET applications requiring time series data. The HYPR technique involves the creation of a composite image from the entire time series. The individual time frames then provide the basis for weighting matrices of the composite. The signal-to-noise ratio (SNR) of the individual time frames can be dramatically improved using the high SNR of ...[more]
Similar Datasets
Project description:BackgroundAccurate quantification of PET studies depends on the spatial resolution of the PET data. The commonly limited PET resolution results in partial volume effects (PVE). Iterative deconvolution methods (IDM) have been proposed as a means to correct for PVE. IDM improves spatial resolution of PET studies without the need for structural information (e.g. MR scans). On the other hand, deconvolution also increases noise, which results in lower signal-to-noise ratios (SNR). The aim of this study was to implement IDM in combination with HighlY constrained back-PRojection (HYPR) denoising to mitigate poor SNR properties of conventional IDM.MethodsAn anthropomorphic Hoffman brain phantom was filled with an [18F]FDG solution of ~25 kBq mL-1 and scanned for 30 min on a Philips Ingenuity TF PET/CT scanner (Philips, Cleveland, USA) using a dynamic brain protocol with various frame durations ranging from 10 to 300 s. Van Cittert IDM was used for PVC of the scans. In addition, HYPR was used to improve SNR of the dynamic PET images, applying it both before and/or after IDM. The Hoffman phantom dataset was used to optimise IDM parameters (number of iterations, type of algorithm, with/without HYPR) and the order of HYPR implementation based on the best average agreement of measured and actual activity concentrations in the regions. Next, dynamic [11C]flumazenil (five healthy subjects) and [11C]PIB (four healthy subjects and four patients with Alzheimer's disease) scans were used to assess the impact of IDM with and without HYPR on plasma input-derived distribution volumes (V T) across various regions of the brain.ResultsIn the case of [11C]flumazenil scans, Hypr-IDM-Hypr showed an increase of 5 to 20% in the regional V T whereas a 0 to 10% increase or decrease was seen in the case of [11C]PIB depending on the volume of interest or type of subject (healthy or patient). References for these comparisons were the V Ts from the PVE-uncorrected scans.ConclusionsIDM improved quantitative accuracy of measured activity concentrations. Moreover, the use of IDM in combination with HYPR (Hypr-IDM-Hypr) was able to correct for PVE without increasing noise.
| S-EPMC5400775 | biostudies-literature
Project description:Highly constrained backprojection-local reconstruction (HYPR-LR) has made a dramatic impact on magnetic resonance angiography (MRA) and shows promise for positron emission tomography (PET) because of the improvements in the signal-to-noise ratio (SNR) it provides dynamic images. For PET in particular, HYPR-LR could improve kinetic analysis methods that are sensitive to noise. In this work, the authors closely examine the performance of HYPR-LR in the context of kinetic analysis, they develop an implementation of the algorithm that can be tailored to specific PET imaging tasks to minimize bias and maximize improvement in variance, and they provide a framework for validating the use of HYPR-LR processing for a particular imaging task.HYPR-LR can introduce errors into non sparse PET studies that might bias kinetic parameter estimates. An implementation of HYPR-LR is proposed that uses multiple temporally summed composite images that are formed based on the kinetics of the tracer being studied (HYPR-LR-MC). The effects of HYPR-LR-MC and of HYPR-LR using a full composite formed with all the frames in the study (HYPR-LR-FC) on the kinetic analysis of Pittsburgh compound-B ([11C]-PIB) are studied. HYPR-LR processing is compared to spatial smoothing. HYPR-LR processing was evaluated using both simulated and human studies. Nondisplaceable binding potential (BP(ND)) parametric images were generated from fifty noise realizations of the same numerical phantom and eight [(11)C]-PIB positive human scans before and after HYPR-LR processing or smoothing using the reference region Logan graphical method and receptor parametric mapping (RPM2). The bias and coefficient of variation in the frontal and parietal cortex in the simulated parametric images were calculated to evaluate the absolute performance of HYPR-LR processing. Bias in the human data was evaluated by comparing parametric image BP(ND) values averaged over large regions of interest (ROIs) to Logan estimates of the BP(ND) from TACs averaged over the same ROIs. Variance was assessed qualitatively in the parametric images and semiquantitatively by studying the correlation between voxel BP(ND) estimates from Logan analysis and RPM2.Both the simulated and human data show that HYPR-LR-FC overestimates BP(ND) values in regions of high [(11)C]-PIB uptake. HYPR-LR-MC virtually eliminates this bias. Both implementations of HYPR-LR reduce variance in the parametric images generated with both Logan analysis and RPM2, and HYPR-LR-FC provides a greater reduction in variance. This reduction in variance nearly eliminates the noise-dependent Logan bias. The variance reduction is greater for the Logan method, particularly for HYPR-LR-MC, and the variance in the resulting Logan images is comparable to that in the RPM2 images. HYPR-LR processing compares favorably with spatial smoothing, particularly when the data are analyzed with the Logan method, as it provides a reduction in variance with no loss of spatial resolution.HYPR-LR processing shows significant potential for reducing variance in parametric images, and can eliminate the noise-dependent Logan bias. HYPR-LR-FC processing provides the greatest reduction in variance but introduces a positive bias into the BP(ND) of high-uptake border regions. The proposed method for forming HYPR composite images, HYPR-LR-MC, eliminates this bias at the cost of less variance reduction.
| S-EPMC3371076 | biostudies-other
Project description:Recently, spectral methods have been extensively used in the processing of 3D meshes. They usually take advantage of some unique properties that the eigenvalues and the eigenvectors of the decomposed Laplacian matrix have. However, despite their superior behavior and performance, they suffer from computational complexity, especially while the number of vertices of the model increases. In this work, we suggest the use of a fast and efficient spectral processing approach applied to dense static and dynamic 3D meshes, which can be ideally suited for real-time denoising and compression applications. To increase the computational efficiency of the method, we exploit potential spectral coherence between adjacent parts of a mesh and then we apply an orthogonal iteration approach for the tracking of the graph Laplacian eigenspaces. Additionally, we present a dynamic version that automatically identifies the optimal subspace size that satisfies a given reconstruction quality threshold. In this way, we overcome the problem of the perceptual distortions, due to the fixed number of subspace sizes that is used for all the separated parts individually. Extensive simulations carried out using different 3D models in different use cases (i.e., compression and denoising), showed that the proposed approach is very fast, especially in comparison with the SVD based spectral processing approaches, while at the same time the quality of the reconstructed models is of similar or even better reconstruction quality. The experimental analysis also showed that the proposed approach could also be used by other denoising methods as a preprocessing step, in order to optimize the reconstruction quality of their results and decrease their computational complexity since they need fewer iterations to converge.
| S-EPMC8321066 | biostudies-literature
Project description:PurposeImage quality of positron emission tomography (PET) is limited by various physical degradation factors. Our study aims to perform PET image denoising by utilizing prior information from the same patient. The proposed method is based on unsupervised deep learning, where no training pairs are needed.MethodsIn this method, the prior high-quality image from the patient was employed as the network input and the noisy PET image itself was treated as the training label. Constrained by the network structure and the prior image input, the network was trained to learn the intrinsic structure information from the noisy image and output a restored PET image. To validate the performance of the proposed method, a computer simulation study based on the BrainWeb phantom was first performed. A 68Ga-PRGD2 PET/CT dataset containing 10 patients and a 18F-FDG PET/MR dataset containing 30 patients were later on used for clinical data evaluation. The Gaussian, non-local mean (NLM) using CT/MR image as priors, BM4D, and Deep Decoder methods were included as reference methods. The contrast-to-noise ratio (CNR) improvements were used to rank different methods based on Wilcoxon signed-rank test.ResultsFor the simulation study, contrast recovery coefficient (CRC) vs. standard deviation (STD) curves showed that the proposed method achieved the best performance regarding the bias-variance tradeoff. For the clinical PET/CT dataset, the proposed method achieved the highest CNR improvement ratio (53.35% ± 21.78%), compared with the Gaussian (12.64% ± 6.15%, P = 0.002), NLM guided by CT (24.35% ± 16.30%, P = 0.002), BM4D (38.31% ± 20.26%, P = 0.002), and Deep Decoder (41.67% ± 22.28%, P = 0.002) methods. For the clinical PET/MR dataset, the CNR improvement ratio of the proposed method achieved 46.80% ± 25.23%, higher than the Gaussian (18.16% ± 10.02%, P < 0.0001), NLM guided by MR (25.36% ± 19.48%, P < 0.0001), BM4D (37.02% ± 21.38%, P < 0.0001), and Deep Decoder (30.03% ± 20.64%, P < 0.0001) methods. Restored images for all the datasets demonstrate that the proposed method can effectively smooth out the noise while recovering image details.ConclusionThe proposed unsupervised deep learning framework provides excellent image restoration effects, outperforming the Gaussian, NLM methods, BM4D, and Deep Decoder methods.
| S-EPMC7814987 | biostudies-literature
Project description:This study describes a HighlY constrained backPRojection (HYPR) image processing method for the reduction of image noise in low tube current time-resolved CT myocardial perfusion scans. The effect of this method on myocardial time-attenuation curve noise and fidelity is evaluated in an animal model, using varying levels of tube current.CT perfusion scans of four healthy pigs (42-59 kg) were acquired at 500, 250, 100, 50, 25, and 10 mA on a 64-slice scanner (4 cm axial coverage, 120 kV, 0.4 s∕rotation, 50 s scan duration). For each scan a sequence of ECG-gated images centered on 75% R-R was reconstructed using short-scan filtered back projection (FBP). HYPR processing was applied to the scans acquired at less than 500 mA using parameters designed to maintain the voxel noise level in the 500-mA FBP images. The processing method generates a series of composite images by averaging over a sliding time window and then multiplies the composite images by weighting images to restore temporal fidelity to the image sequence. HYPR voxel noise relative to FBP noise was measured in AHA myocardial segment numbers 1, 5, 6, and 7 at each mA. To quantify the agreement between HYPR and FBP time-attenuation curves (TACs), Bland-Altman analysis was performed on TACs measured in full myocardial segments. The relative degree of TAC fluctuation in smaller subvolumes was quantified by calculating the root mean square deviation of a TAC about the gamma variate curve fit to the TAC data.HYPR image sequences were produced using 2, 7, and 20 beat composite windows for the 250, 100, and 50 mA scans, respectively. At 25 and 10 mA, all available beats were used in the composite (41-60; average 50). A 7-voxel-wide 3D cubic filter kernel was used to form weighting images. The average ratio of HYPR voxel noise to 500-mA FBP voxel noise was 1.06, 1.10, 0.97, 1.11, and 2.15 for HYPR scans at 250, 100, 50, 25, and 10 mA. The average limits-of-agreement between HYPR and FBP TAC values measured 0.02+∕-0.91, 0.04+∕-1.92, 0.19+∕-1.59, 1.13+∕-4.22, and 1.07+∕-6.37 HU (mean difference +∕-1.96 SD). The HYPR image subvolume that yielded a fixed level of TAC fluctuations was smaller, on average, than the FBP subvolume determined at the same mA.HYPR processing is a feasible method for generating low noise myocardial perfusion data from a low-mA time-resolved CT myocardial perfusion scan. The method is applicable to current clinical scanners and uses conventional image reconstructions as input data.
| S-EPMC3537716 | biostudies-other
Project description: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.
| S-EPMC8638912 | biostudies-literature
Project description:BackgroundWith a constantly increasing number of diagnostic images performed each year, Artificial Intelligence (AI) denoising methods offer an opportunity to respond to the growing demand. However, it may affect information in the image in an unknown manner. This study quantifies the effect of AI-based denoising on FDG PET textural information in comparison to a convolution with a standard gaussian postfilter (EARL1).MethodsThe study was carried out on 113 patients who underwent a digital FDG PET/CT (VEREOS, Philips Healthcare). 101 FDG avid lesions were segmented semi-automatically by a nuclear medicine physician. VOIs in the liver and lung as reference organs were contoured. PET textural features were extracted with pyradiomics. Texture features from AI denoised and EARL1 versus original PET images were compared with a Concordance Correlation Coefficient (CCC). Features with CCC values ≥ 0.85 threshold were considered concordant. Scatter plots of variable pairs with R2 coefficients of the more relevant features were computed. A Wilcoxon signed rank test to compare the absolute values between AI denoised and original images was performed.ResultsThe ratio of concordant features was 90/104 (86.5%) in AI denoised versus 46/104 (44.2%) with EARL1 denoising. In the reference organs, the concordant ratio for AI and EARL1 denoised images was low, respectively 12/104 (11.5%) and 7/104 (6.7%) in the liver, 26/104 (25%) and 24/104 (23.1%) in the lung. SUVpeak was stable after the application of both algorithms in comparison to SUVmax. Scatter plots of variable pairs showed that AI filtering affected more lower versus high intensity regions unlike EARL1 gaussian post filters, affecting both in a similar way. In lesions, the majority of texture features 79/100 (79%) were significantly (p<0.05) different between AI denoised and original PET images.ConclusionsApplying an AI-based denoising on FDG PET images maintains most of the lesion's texture information in contrast to EARL1-compatible Gaussian filter. Predictive features of a trained model could be thus the same, however with an adapted threshold. Artificial intelligence based denoising in PET is a very promising approach as it adapts the denoising in function of the tissue type, preserving information where it should.
| S-EPMC8421788 | biostudies-literature
Project description:BackgroundAdvancements in hybrid positron emission tomography-magnetic resonance (PET-MR) systems allow for combining the advantages of each modality. Integrating information from MRI and PET can be valuable for diagnosing and treating neurological disorders. However, combining diffusion MRI (dMRI) and PET data, which provide highly complementary information, has rarely been exploited in image post-processing. dMRI has the ability to investigate the white matter pathways of the brain through fibre tractography, which enables comprehensive mapping of the brain connection networks (the "connectome"). Novel methods are required to combine information present in the connectome and PET to increase the full potential of PET-MRI.MethodsWe developed a CONNectome-based Non-Local Means (CONN-NLM) filter to exploit synergies between dMRI-derived structural connectivity and PET intensity information to denoise PET images. PET-MR data are parcelled into a number of regions based on a brain atlas, and the inter-regional structural connectivity is calculated based on dMRI fibre-tracking. The CONN-NLM filter is then implemented as a post-reconstruction filter by combining the nonlocal means filter and a connectivity-based cortical smoothing. The effect of this approach is to weight voxels with similar PET intensity and highly connected voxels higher when computing the weighted-average to perform more informative denoising. The proposed method was first evaluated using a novel computer phantom framework to simulate realistic hybrid PET-MR images with different lesion scenarios. CONN-NLM was further assessed with clinical dMRI and tau PET examples.ResultsThe results showed that CONN-NLM has the capacity to improve the overall PET image quality by reducing noise while preserving lesion contrasts, and it outperformed a range of filters that did not use dMRI information. The simulations demonstrate that CONN-NLM can handle various lesion contrasts consistently, as well as lesions with different levels of inter-connectivity.ConclusionCONN-NLM has unique advantages of providing more informative and accurate PET smoothing by adding complementary structural connectivity information from dMRI, representing a new avenue to exploit synergies between MRI and PET.
| S-EPMC9197079 | biostudies-literature
Project description:A new dynamic mode decomposition (DMD) method is introduced for simultaneous system identification and denoising in conjunction with the adoption of an extended Kalman filter algorithm. The present paper explains the extended-Kalman-filter-based DMD (EKFDMD) algorithm which is an online algorithm for dataset for a small number of degree of freedom (DoF). It also illustrates that EKFDMD requires significant numerical resources for many-degree-of-freedom (many-DoF) problems and that the combination with truncated proper orthogonal decomposition (trPOD) helps us to apply the EKFDMD algorithm to many-DoF problems, though it prevents the algorithm from being fully online. The numerical experiments of a noisy dataset with a small number of DoFs illustrate that EKFDMD can estimate eigenvalues better than or as well as the existing algorithms, whereas EKFDMD can also denoise the original dataset online. In particular, EKFDMD performs better than existing algorithms for the case in which system noise is present. The EKFDMD with trPOD, which unfortunately is not fully online, can be successfully applied to many-DoF problems, including a fluid-problem example, and the results reveal the superior performance of system identification and denoising.
| S-EPMC6383872 | biostudies-literature
Project description:PURPOSE:PET and SPECT voxel kinetics are highly noised. To our knowledge, no study has determined the effect of denoising on the ability to detect differences in binding at the voxel level using Statistical Parametric Mapping (SPM). METHODS:In the present study, groups of subject-images with a 10%- and 20%- difference in binding of [123I]iomazenil (IMZ) were simulated. They were denoised with Factor Analysis (FA). Parametric images of binding potential (BPND) were produced with the simplified reference tissue model (SRTM) and the Logan non-invasive graphical analysis (LNIGA) and analyzed using SPM to detect group differences. FA was also applied to [123I]IMZ and [11C]flumazenil (FMZ) clinical images (n = 4) and the variance of BPND was evaluated. RESULTS:Estimations from FA-denoised simulated images provided a more favorable bias-precision profile in SRTM and LNIGA quantification. Simulated differences were detected in a higher number of voxels when denoised simulated images were used for voxel-wise estimations, compared to quantification on raw simulated images. Variability of voxel-wise binding estimations on denoised clinical SPECT and PET images was also significantly diminished. CONCLUSION:In conclusion, noise removal from dynamic brain SPECT and PET images may optimize voxel-wise BPND estimations and detection of biological differences using SPM.
| S-EPMC6124809 | biostudies-literature