PET Attenuation Map Research Articles

Generating PET Attenuation Maps via Sim2Real Deep Learning-Based Tissue Composition Estimation Combined with MLACF.

Deep learning (DL) has recently attracted attention for data processing in positron emission tomography (PET). Attenuation correction (AC) without computed tomography (CT) data is one of the interests. Here, we present, to our knowledge, the first attempt to generate an attenuation map of the human head via Sim2Real DL-based tissue composition estimation from model training using only the simulated PET dataset. The DL model accepts a two-dimensional non-attenuation-corrected PET image as input and outputs a four-channel tissue-composition map of soft tissue, bone, cavity, and background. Then, an attenuation map is generated by a linear combination of the tissue composition maps and, finally, used as input for scatter+randomestimation and as an initial estimate for attenuation map reconstruction by the maximum likelihood attenuation correction factor (MLACF), i.e., the DL estimate is refined by the MLACF. Preliminary results using clinical brain PET data showed that the proposed DL model tended to estimate anatomical details inaccurately, especially in the neck-side slices. However, it succeeded in estimating overall anatomical structures, and the PET quantitative accuracy with DL-based AC was comparable to that with CT-based AC. Thus, the proposed DL-based approach combined with the MLACF is also a promising CT-less AC approach.

Eliminating CT radiation for clinical PET examination using deep learning

Clinical PET/CT examinations rely on CT modality for anatomical localization and attenuation correction of the PET data. However, the use of CT significantly increases the risk of ionizing radiation exposure for patients. We propose a deep learning framework to learn the relationship mapping between attenuation corrected (AC) PET and non-attenuation corrected (NAC) PET images to estimate PET attenuation maps and generate pseudo-CT images for medical observation. In this study, 5760, 1608 and 1351 pairs of transverse PET-CT slices were used as the training, validation, and testing sets, respectively, to implement the proposed framework. A pix2pix model was adopted to predict AC PET images from NAC PET images, which allowed the calculation of PET attenuation maps (µ-maps). The same model was then applied to generate realistic CT images from the calculated µ-maps. The quality of predicted AC PET and CT was assessed using normalized root mean square error (NRMSE), peak signal-to-noise ratio (PSNR), structural similarity index (SSIM) and Pearson correlation coefficient (PCC). Relative to true AC PET, the synthetic AC PET achieved superior quantitative performances with 2.20 ± 1.17% NRMSE, 34.03 ± 4.73 dB PSNR, 97.90 ± 1.22% SSIM and 98.45 ± 1.31% PCC. The synthetic CT and synthetic AC PET images were deemed acceptable by radiologists who rated the images, as they provided sufficient anatomical and functional information, respectively. This work demonstrates that the proposed deep learning framework is a promising method in clinical applications, such as radiotherapy and low-dose imaging.

MRI-guided attenuation correction in torso PET/MRI: Assessment of segmentation-, atlas-, and deep learning-based approaches in the presence of outliers.

PurposeWe compare the performance of three commonly used MRI‐guided attenuation correction approaches in torso PET/MRI, namely segmentation‐, atlas‐, and deep learning‐based algorithms.MethodsTwenty‐five co‐registered torso 18F‐FDG PET/CT and PET/MR images were enrolled. PET attenuation maps were generated from in‐phase Dixon MRI using a three‐tissue class segmentation‐based approach (soft‐tissue, lung, and background air), voxel‐wise weighting atlas‐based approach, and a residual convolutional neural network. The bias in standardized uptake value (SUV) was calculated for each approach considering CT‐based attenuation corrected PET images as reference. In addition to the overall performance assessment of these approaches, the primary focus of this work was on recognizing the origins of potential outliers, notably body truncation, metal‐artifacts, abnormal anatomy, and small malignant lesions in the lungs.ResultsThe deep learning approach outperformed both atlas‐ and segmentation‐based methods resulting in less than 4% SUV bias across 25 patients compared to the segmentation‐based method with up to 20% SUV bias in bony structures and the atlas‐based method with 9% bias in the lung. The deep learning‐based method exhibited superior performance. Yet, in case of sever truncation and metallic‐artifacts in the input MRI, this approach was outperformed by the atlas‐based method, exhibiting suboptimal performance in the affected regions. Conversely, for abnormal anatomies, such as a patient presenting with one lung or small malignant lesion in the lung, the deep learning algorithm exhibited promising performance compared to other methods.ConclusionThe deep learning‐based method provides promising outcome for synthetic CT generation from MRI. However, metal‐artifact and body truncation should be specifically addressed.

Assessment of motion correction in dynamic rubidium-82 cardiac PET with and without frame-by-frame adjustment of attenuation maps for calculation of myocardial blood flow

BackgroundPatient motion during pharmacological stressing can have substantial impact on myocardial blood flow (MBF) estimated from dynamic PET. This work evaluated a motion correction algorithm with and without adjustment of the PET attenuation map. MethodsFrame-by-frame motion correction was performed by three users on 30 rubidium-82 studies. Data were divided equally into three groups of motion severity [mild (M1), moderate (M2) and severe (M3)]. MBF data were compared for non-motion corrected (NC), motion-corrected-only (MC) and with adjustment of the attenuation map (MCAC). Percentage differences of MBF were calculated in the coronary territories and 17-segment polar plots. Polar plots of spill-over were also generated from the data. ResultsMedian differences of 23% were seen in the RCA and 18% for the LAD in the M3 category for MC vs NC images. Differences for MCAC vs MC images were considerably smaller and typically < 10%. Spill-over plots for MC and MCAC were notably more uniform compared with NC images. ConclusionMotion correction for dynamic rubidium data is desirable for future MBF software updates. Adjustment of the PET attenuation map results in only marginal differences and therefore is unlikely to be an essential requirement. Assessing the uniformity of spill-over plots is a useful visual aid for verifying motion correction techniques.

An Aristotelian View on MR-Based Attenuation Correction (ARISTOMRAC): Combining the Four Elements

MR-based attenuation correction (MRAC) is important for accurate quantification of the uptake of PET tracers in combined PET/MR scanners. However, current techniques for MRAC usually require multiple acquisitions or complex post-processing to discriminate the different tissues. Inspired by the ancient Greeks, who believed that matter was made of the combination of four elements (earth, water, air and fire), we formulated a multi-component Magnetic Resonance (MR) Fingerprinting framework, where every voxel was considered a weighted combination of four base elements: bone, water, air and fat. We named our approach Aristotelian MR based attenuation correction (ARISTOMRAC). We used a 3D radial acquisition scheme at 1.5T, acquiring a transient-state spoiled acquisition with variable flip angles and echo times (TE), with the shortest TEs being ultra-short echo times (UTE). We simulated a multi-tissue MR signal model using the Bloch equations and used dictionary matching to extract tissue fraction maps for bone water and fat, while air fractions were obtained by thresholding the UTE parts of our acquisitions at higher spatial resolution. Compared to previous methods for MR-based Attenuation Correction (MRAC), our approach used a full multi-component signal model, including multiple tissues per voxel. For this reason, rather than reconstructing high resolutions images, MR data can be acquired more efficiently, directly at the resolution needed for PET attenuation maps. The ARISTOMRAC method allows to accurately estimate the air, water, bone and fat fractions (Concordance Correlation Coefficient = 0.81/0.91/0.98 for bone, water and fat respectively). Attenuation maps could be obtained in the head and neck with a single 1-minute acquisition.

Generation of PET Attenuation Map for Whole-Body Time-of-Flight 18F-FDG PET/MRI Using a Deep Neural Network Trained with Simultaneously Reconstructed Activity and Attenuation Maps.

We propose a new deep learning-based approach to provide more accurate whole-body PET/MRI attenuation correction than is possible with the Dixon-based 4-segment method. We use activity and attenuation maps estimated using the maximum-likelihood reconstruction of activity and attenuation (MLAA) algorithm as inputs to a convolutional neural network (CNN) to learn a CT-derived attenuation map. Methods: The whole-body 18F-FDG PET/CT scan data of 100 cancer patients (38 men and 62 women; age, 57.3 ± 14.1 y) were retrospectively used for training and testing the CNN. A modified U-net was trained to predict a CT-derived μ-map (μ-CT) from the MLAA-generated activity distribution (λ-MLAA) and μ-map (μ-MLAA). We used 1.3 million patches derived from 60 patients' data for training the CNN, data of 20 others were used as a validation set to prevent overfitting, and the data of the other 20 were used as a test set for the CNN performance analysis. The attenuation maps generated using the proposed method (μ-CNN), μ-MLAA, and 4-segment method (μ-segment) were compared with the μ-CT, a ground truth. We also compared the voxelwise correlation between the activity images reconstructed using ordered-subset expectation maximization with the μ-maps, and the SUVs of primary and metastatic bone lesions obtained by drawing regions of interest on the activity images. Results: The CNN generates less noisy attenuation maps and achieves better bone identification than MLAA. The average Dice similarity coefficient for bone regions between μ-CNN and μ-CT was 0.77, which was significantly higher than that between μ-MLAA and μ-CT (0.36). Also, the CNN result showed the best pixel-by-pixel correlation with the CT-based results and remarkably reduced differences in activity maps in comparison to CT-based attenuation correction. Conclusion: The proposed deep neural network produced a more reliable attenuation map for 511-keV photons than the 4-segment method currently used in whole-body PET/MRI studies.

Synthesis of Patient-Specific Transmission Data for PET Attenuation Correction for PET/MRI Neuroimaging Using a Convolutional Neural Network.

Attenuation correction is a notable challenge associated with simultaneous PET/MRI, particularly in neuroimaging, where sharp boundaries between air and bone volumes exist. This challenge leads to concerns about the visual and, more specifically, quantitative accuracy of PET reconstructions for data obtained with PET/MRI. Recently developed techniques can synthesize attenuation maps using only MRI data and are likely adequate for clinical use; however, little work has been conducted to assess their suitability for the dynamic PET studies frequently used in research to derive physiologic information such as the binding potential of neuroreceptors in a region. At the same time, existing PET/MRI attenuation correction methods are predicated on synthesizing CT data, which is not ideal, as CT data are acquired with much lower-energy photons than PET data and thus do not optimally reflect the PET attenuation map. Methods: We trained a convolutional neural network to generate patient-specific transmission data from T1-weighted MRI. Using the trained network, we generated transmission data for a testing set comprising 11 subjects scanned with 11C-labeled N-[2-]4-(2-methoxyphenyl)-1-piperazinyl]ethyl]-N-(2-pyridinyl)cyclohexanecarboxamide) (11C-WAY-100635) and 10 subjects scanned with 11C-labeled 3-amino-4-(2-dimethylaminomethyl-phenylsulfanyl)benzonitrile (11C-DASB). We assessed both static and dynamic reconstructions. For dynamic PET data, we report differences in both the nondisplaceable and the free binding potential for 11C-WAY-100635 and distribution volume for 11C-DASB. Results: The mean bias for generated transmission data was -1.06% ± 0.81%. Global biases in static PET uptake were -0.49% ± 1.7%, and -1.52% ± 0.73% for 11C-WAY-100635 and 11C-DASB, respectively. Conclusion: Our neural network approach is capable of synthesizing patient-specific transmission data with sufficient accuracy for both static and dynamic PET studies.

Direct Reconstruction of CT-based Attenuation Correction Images for PET with Cluster-Based Penalties.

Extremely low-dose CT acquisitions used for PET attenuation correction have high levels of noise and potential bias artifacts due to photon starvation. This work explores the use of a priori knowledge for iterative image reconstruction of the CT-based attenuation map. We investigate a maximum a posteriori framework with cluster-based multinomial penalty for direct iterative coordinate decent (dICD) reconstruction of the PET attenuation map. The objective function for direct iterative attenuation map reconstruction used a Poisson log-likelihood data fit term and evaluated two image penalty terms of spatial and mixture distributions. The spatial regularization is based on a quadratic penalty. For the mixture penalty, we assumed that the attenuation map may consist of four material clusters: air+background, lung, soft tissue, and bone. Using simulated noisy sinogram data, dICD reconstruction was performed with different strengths of the spatial and mixture penalties. The combined spatial and mixture penalties reduced the RMSE by roughly 2 times compared to a weighted least square and filtered backprojection reconstruction of CT images. The combined spatial and mixture penalties resulted in only slightly lower RMSE compared to a spatial quadratic penalty alone. For direct PET attenuation map reconstruction from ultra-low dose CT acquisitions, the combination of spatial and mixture penalties offers regularization of both variance and bias and is a potential method to reconstruct attenuation maps with negligible patient dose. The presented results, using a best-case histogram suggest that the mixture penalty does not offer a substantive benefit over conventional quadratic regularization and diminishes enthusiasm for exploring future application of the mixture penalty.

Feasibility of using respiration-averaged MR images for attenuation correction of cardiac PET/MR imaging.

Cardiac imaging is a promising application for combined PET/MR imaging. However, current MR imaging protocols for whole‐body attenuation correction can produce spatial mismatch between PET and MR‐derived attenuation data owing to a disparity between the two modalities' imaging speeds. We assessed the feasibility of using a respiration‐averaged MR (AMR) method for attenuation correction of cardiac PET data in PET/MR images. First, to demonstrate the feasibility of motion imaging with MR, we used a 3T MR system and a two‐dimensional fast spoiled gradient‐recalled echo (SPGR) sequence to obtain AMR images of a moving phantom. Then, we used the same sequence to obtain AMR images of a patient's thorax under free‐breathing conditions. MR images were converted into PET attenuation maps using a three‐class tissue segmentation method with two sets of predetermined CT numbers, one calculated from the patient‐specific (PS) CT images and the other from a reference group (RG) containing 54 patient CT datasets. The MR‐derived attenuation images were then used for attenuation correction of the cardiac PET data, which were compared to the PET data corrected with average CT (ACT) images. In the myocardium, the voxel‐by‐voxel differences and the differences in mean slice activity between the AMR‐corrected PET data and the ACT‐corrected PET data were found to be small (less than 7%). The use of AMR‐derived attenuation images in place of ACT images for attenuation correction did not affect the summed stress score. These results demonstrate the feasibility of using the proposed SPGR‐based MR imaging protocol to obtain patient AMR images and using those images for cardiac PET attenuation correction. Additional studies with more clinical data are warranted to further evaluate the method.PACS number: 87.57.uk

Attenuation Effects of MR Headphones During Brain PET/MR Studies

PET/MR offers potential advantages over PET/CT that are currently under investigation. One of the challenges of PET/MR is attenuation correction, as there is no simple correlation between MR signal intensity and the attenuation of 511-keV photons detected in PET. Currently, dedicated MR sequences are used to segment voxels into categories that are then assigned a predetermined attenuation coefficient. MR hardware such as the imaging table, coils, and headphones are also sources of attenuation. The purpose of this study was to evaluate the effect of MR-compatible headphones on average activity concentration measured with PET/MR. We also present a practical approach to correct for the attenuation effect of headphones using a CT-derived attenuation map. Phantom studies were performed using a 3-L cylindric phantom containing 55 MBq of (18)F-FDG and water. Images were acquired on a PET/MR device in 2 settings-one with the PET/MR headphones on and one with the headphones off. Phantom images were analyzed to compare activity concentration with headphones on and off. A high-resolution CT and (57)Co transmission scan was obtained to construct a PET attenuation map of the headphones. The resulting attenuation map was registered to the phantom data to evaluate the ability to correct for headphone attenuation. One human subject was scanned to evaluate the clinical impact of headphone attenuation and the accuracy of the proposed correction. Activity concentrations measured in the phantom were reduced by as much as 13.2% with headphones on compared with headphones off. Using the modified attenuation maps that account for attenuation from the headphones resulted in a decrease in the headphone attenuation effect from a maximum of 13.2% to 1.9%. Comparable attenuation effects were observed in the human brain and were similarly reduced with correction using the modified attenuation maps. MR-safe headphones were a source of attenuation on our PET/MR phantom and human studies. Attenuation effects of headphones should be considered and can be corrected during quantitative brain PET/MR studies.

A generalized method of converting CT image to PET linear attenuation coefficient distribution in PET/CT imaging

The accuracy of attenuation correction in positron emission tomography scanners depends mainly on deriving the reliable 511-keV linear attenuation coefficient distribution in the scanned objects. In the PET/CT system, the linear attenuation distribution is usually obtained from the intensities of the CT image. However, the intensities of the CT image relate to the attenuation of photons in an energy range of 40 keV–140 keV. Before implementing PET attenuation correction, the intensities of CT images must be transformed into the PET 511-keV linear attenuation coefficients. However, the CT scan parameters can affect the effective energy of CT X-ray photons and thus affect the intensities of the CT image. Therefore, for PET/CT attenuation correction, it is crucial to determine the conversion curve with a given set of CT scan parameters and convert the CT image into a PET linear attenuation coefficient distribution. A generalized method is proposed for converting a CT image into a PET linear attenuation coefficient distribution. Instead of some parameter-dependent phantom calibration experiments, the conversion curve is calculated directly by employing the consistency conditions to yield the most consistent attenuation map with the measured PET data. The method is evaluated with phantom experiments and small animal experiments. In phantom studies, the estimated conversion curve fits the true attenuation coefficients accurately, and accurate PET attenuation maps are obtained by the estimated conversion curves and provide nearly the same correction results as the true attenuation map. In small animal studies, a more complicated attenuation distribution of the mouse is obtained successfully to remove the attenuation artifact and improve the PET image contrast efficiently.

Direct Reconstruction of CT-based Attenuation Correction Images for PET with Cluster-Based Penalties.

Extremely low-dose CT acquisitions for the purpose of PET attenuation correction will have a high level of noise and biasing artifacts due to factors such as photon starvation. This work explores a priori knowledge appropriate for CT iterative image reconstruction for PET attenuation correction. We investigate the maximum a posteriori (MAP) framework with cluster-based, multinomial priors for the direct reconstruction of the PET attenuation map. The objective function for direct iterative attenuation map reconstruction was modeled as a Poisson log-likelihood with prior terms consisting of quadratic (Q) and mixture (M) distributions. The attenuation map is assumed to have values in 4 clusters: air+background, lung, soft tissue, and bone. Under this assumption, the MP was a mixture probability density function consisting of one exponential and three Gaussian distributions. The relative proportion of each cluster was jointly estimated during each voxel update of direct iterative coordinate decent (dICD) method. Noise-free data were generated from NCAT phantom and Poisson noise was added. Reconstruction with FBP (ramp filter) was performed on the noise-free (ground truth) and noisy data. For the noisy data, dICD reconstruction was performed with the combination of different prior strength parameters (β and γ) of Q- and M-penalties. The combined quadratic and mixture penalties reduces the RMSE by 18.7% compared to post-smoothed iterative reconstruction and only 0.7% compared to quadratic alone. For direct PET attenuation map reconstruction from ultra-low dose CT acquisitions, the combination of quadratic and mixture priors offers regularization of both variance and bias and is a potential method to derive attenuation maps with negligible patient dose. However, the small improvement in quantitative accuracy relative to the substantial increase in algorithm complexity does not currently justify the use of mixture-based PET attenuation priors for reconstruction of CT images for PET attenuation correction.

Effect of MR contrast agents on quantitative accuracy of PET in combined whole-body PET/MR imaging

Clinical PET/MR acquisition protocols entail the use of MR contrast agents (MRCA) that could potentially affect PET quantification following MR-based attenuation correction (AC). We assessed the effect of oral and intravenous (IV) MRCA on PET quantification in PET/MR imaging. We employed two MRCA: Lumirem (oral) and Gadovist (IV). First, we determined their reference PET attenuation values using a PET transmission scan (ECAT-EXACT HR+, Siemens) and a CT scan (PET/CT Biograph 16 HI-REZ, Siemens). Second, we evaluated the attenuation of PET signals in the presence of MRCA. Phantoms were filled with clinically relevant concentrations of MRCA in a background of water and (18)F-fluoride, and imaged using a PET/CT scanner (Biograph 16 HI-REZ, Siemens) and a PET/MR scanner (Biograph mMR, Siemens). Third, we investigated the effect of clinically relevant volumes of MRCA on MR-based AC using human pilot data: a patient study employing Gadovist (IV) and a volunteer study employing two different oral MRCA (Lumirem and pineapple juice). MR-based attenuation maps were calculated following Dixon-based fat-water segmentation and an external atlas-based and pattern recognition (AT&PR) algorithm. IV and oral MRCA in clinically relevant concentrations were found to have PET attenuation values similar to those of water. The phantom experiments showed that under clinical conditions IV and oral MRCA did not yield additional attenuation of PET emission signals. Patient scans showed that PET attenuation maps are not biased after the administration of IV MRCA but may be biased, however, after ingestion of iron oxide-based oral MRCA when segmentation-based AC algorithms are used. Alternative AC algorithms, such as AT&PR, or alternative oral contrast agents, such as pineapple juice, can yield unbiased attenuation maps. In clinical PET/MR scenarios MRCA are not expected to lead to markedly increased attenuation of the PET emission signals. MR-based attenuation maps may be biased by oral iron oxide-based MRCA unless advanced AC algorithms are used.

Application of discrete data consistency conditions for selecting regularization parameters in PET attenuation map reconstruction

Simultaneous emission and transmission measurement is appealing in PET due to the matching of geometrical conditions between emission and transmission and reduced acquisition time for the study. A potential problem remains: when transmission statistics are low, attenuation correction could be very noisy. Although noise in the attenuation map can be controlled through regularization during statistical reconstruction, the selection of regularization parameters is usually empirical. In this paper, we investigate the use of discrete data consistency conditions (DDCC) to optimally select one or two regularization parameters. The advantages of the method are that the reconstructed attenuation map is consistent with the emission data and that it accounts for particularity in the emission reconstruction algorithm and acquisition geometry. The methodology is validated using a computer-generated whole-body phantom for both emission and transmission, neglecting random events and scattered radiation. MAP-TR was used for attenuation map reconstruction, while 3D OS-EM is used for estimating the emission image. The estimation of regularization parameters depends on the resolution of the emission image controlled by the number of iterations in OS-EM. The computer simulation shows that, on one hand, DDCC regularized attenuation map reduces propagation of the transmission scan noise to the emission image, while on the other hand DDCC prevents excessive attenuation map smoothing that could result in resolution mismatch artefacts between emission and transmission.