Lung attenuation coefficient estimation using Maximum Likelihood reconstruction of attenuation and activity for PET/MR attenuation correction

We propose a deep learning-based data-driven respiratory phase-matched gated-PET attenuation correction (AC) method that does not need a gated-CT. The proposed method is a multi-step process that consists of data-driven respiratory gating, gated attenuation map estimation using maximum-likelihood reconstruction of attenuation and activity (MLAA) algorithm, and enhancement of the gated attenuation maps using convolutional neural network (CNN). The gated MLAA attenuation maps enhanced by the CNN allowed for the phase-matched AC of gated-PET images. We conducted a non-rigid registration of the gated-PET images to generate motion-free PET images. We trained the CNN by conducting a 3D patch-based learning with 80 oncologic whole-body 18F-fluorodeoxyglucose (18F-FDG) PET/CT scan data and applied it to seven regional PET/CT scans that cover the lower lung and upper liver. We investigated the impact of the proposed respiratory phase-matched AC of PET without utilizing CT on tumor size and standard uptake value (SUV) assessment, and PET image quality (%STD). The attenuation corrected gated and motion-free PET images generated using the proposed method yielded sharper organ boundaries and better noise characteristics than conventional gated and ungated PET images. A banana artifact observed in a phase-mismatched CT-based AC was not observed in the proposed approach. By employing the proposed method, the size of tumor was reduced by 12.3% and SUV90% was increased by 13.3% in tumors with larger movements than 5 mm. %STD of liver uptake was reduced by 11.1%. The deep learning-based data-driven respiratory phase-matched AC method improved the PET image quality and reduced the motion artifacts.

Accurate quantification of the activity distribution in positron emission tomography (PET) mandates attenuation correction (AC). Unlike in PET/CT, AC in PET/MR is, however, challenging, since information about the attenuation properties of the patient tissue distribution is not available directly. Standard MR-based AC (MRAC) does not account for the presence of bone and, thus, yields an underestimation of the activity distribution. We propose an algorithm to simultaneously reconstruct the activity and attenuation distribution using MR images as anatomical prior information for non time-of-flight PET/MR. The proposed algorithm is an extension of the existing maximum-likelihood reconstruction of attenuation and activity (MLAA). The MR images are used to obtain an initial attenuation map and to derive voxel-dependent expectations on the attenuation coefficients. These expectations are modeled using pre-defined attenuation values and Gaussian-like probability functions. An iterative reconstruction scheme incorporating the prior information on the attenuation coefficients is used to update attenuation and activity distribution in an alternating manner. The algorithm, called MR-MLAA, is evaluated for simulated 2D PET data for two patients with artificial lesions in the head region. The proposed algorithm helps recover bone attenuation information. However, for both patients, some misclassifications of air (considered as bone) and bone (considered as air or soft tissue) were observed. Nevertheless, PET quantification in lesions located close to bone tissue is greatly improved when using MR-MLAA. Errors in activity estimation are reduced to ranges of -9% to +1% whereas MRAC yields errors of -22% to -10%. In conclusion, MR-MLAA has the potential to improve quantification in hybrid PET/MR, especially in regions adjacent to dense bone tissue.

Attenuation correction (AC) for both patient and hardware is required for accurate PET quantification. The current clinical value of hybrid PET/MR imaging is limited because MR-based AC (MRAC) does not properly account for bone attenuation and neglects attenuation of flexible hardware components (headphones, flexible body coils). To improve PET quantification for PET/MR, we developed a new reconstruction approach combining ideas of our recently published methods for emission-based patient and hardware AC. The new algorithm, xMR-MLAA, is based on maximum-likelihood reconstruction of attenuation and activity (MLAA), and estimates patient and hardware attenuation in an interleaved manner. The proposed method was evaluated using clinical brain scans acquired with a Siemens Biograph mMR scanner. Conventional T1-weighted MR images were used to derive the patient support as well as anatomical prior information on the patient attenuation distribution. Standard MRAC results in activity underestimation in the brain, caused both by neglecting headphone attenuation and by treating bone as soft tissue during AC. Across three FDG patients included in this study, average brain activity underestimation was 14.8% compared to CTAC. Using xMR- LAA, bone attenuation information could be accurately recovered while preserving air cavities like the paranasal sinuses and the inner ears. Moreover, accurate estimates of the headphone attenuation were obtained. Average brain activity underestimation with xMR- LAA was reduced to 3.1% compared to CTAC.

Quantitative PET image reconstruction requires an accurate map of photon attenuation coefficients (μ-map) in order to correct the PET emission data. Current PET/MR imaging systems use methods based on MR image segmentation with subsequent assignment of empirical attenuation coefficients. In this study we examine the differences in the quantification of 18F-FDG standardized uptake values (SUV) in head and neck cancer, using two different MR imaging sequences for MR-based attenuation correction (MRAC): a zero echo time (ZTE) sequence which can image bone directly (ZTE-MRAC), and a vendor-provided 2-point Dixon sequence that neglects bone (Dixon-MRAC). The μ-maps from each MRAC techniques were compared to CT-based attenuation correction (CTAC) maps. Percent SUV-mean and SUV-max differences in relevant regions of interest (ROIs) were calculated for three patients. Relative to Dixon-MRAC, we observed 15±7% and 14±8% increase of SUV-mean and SUV-max, respectively, when ZTE-based bone information was incorporated in the attenuation map and using Dixon-based attenuation map, respectively. We also observed that use of Dixon-MRAC led to 7±7% and 8±8% underestimation of SUV-mean and SUV-max, respectively, whereas with ZTE-MRAC led to 6±8% and 5±8% higher SUV-mean and SUV-max, respectively, compared to CTAC. This study is the first demonstration of ZTE-based attenuation correction in the head and neck region and compared with CTAC as a gold standard with the goal of improving PET quantitation. The study shows that incorporation of bone information on μ-maps has a significant impact on SUV quantitation in head and neck cancer lesions.

The objective of this study was to introduce and evaluate a method for MR-based attenuation and truncation correction in phantom and patient measurements to improve PET quantification in PET/MR hybrid imaging. The fully MR-based method HUGE (B0 Homogenization using gradient enhancement) provides field-of-view extension in MR imaging, which can be used for truncation correction and improved PET quantification in PET/MR hybrid imaging. The HUGE method in this recent implementation is combined with continuously moving table data acquisition to provide a seamless nontruncated whole-body data set of the outer patient contours to complete the established standard MR-based Dixon-VIBE data for attenuation correction. The method was systematically evaluated in NEMA standard phantom experiments to investigate the impact of HUGE truncation correction on PET quantification. The method was then applied to 24 oncologic patients in whole-body PET/MR hybrid imaging. The impact of MR-based truncation correction with HUGE on PET data was compared to the impact of the established PET-based MLAA algorithm for contour detection. In phantom and in all patient measurements, the standard Dixon-VIBE attenuation correction data show geometric distortions and signal truncations at the edges of the MR imaging field-of-view. In contrast, the Dixon-VIBE-based attenuation correction data additionally extended by applying HUGE shows significantly less distortion and truncations and due to the continuously moving table acquisition robustly provides smooth outer contours of the patient arms. In the investigated patient cases, MLAA frequently showed an overestimation of arm volume and associated artifacts limiting contour detection. When applying HUGE, an average relative increase in SUVmean in patients' lesion of 4.2% and for MLAA of 4.6% were measured, when compared to standard Dixon-VIBE only. In specific lesions maximal differences in SUVmean up to 13% for HUGE and 14% for MLAA were measured. Quantification in truncated regions showed maximal differences up to 40% for both, MLAA and HUGE. Average differences in those regions in SUVmean for HUGE are 13.3% and 14.6% for MLAA. In a patient with I-124 radiotracer PET-based MLAA contour detection completely failed in this specific case, whereas HUGE as MR-based approach provided accurate truncation correction. The HUGE method for truncation correction combined with continuous table movement extends the lateral MR field-of-view and effectively reduces truncations along the outer contours of the patient's arms in whole-body PET/MR imaging. HUGE as a fully MR-based approach is independent of the choice of radiotracer, thus also offering robust truncation correction in patients that are not injected with Fluordesoxyglucose (FDG) as radiotracer. Therefore, this method improves the standard Dixon MR-based attenuation correction and PET image quantification in whole-body PET/MR imaging applications.

PET Attenuation Correction (AC) in small animal imaging is generally performed via transmission (TX) scans. In simultaneous PET/MR scanners, no transmission source is available, and MR-based AC (MRAC) or AC from emission data is necessary. However, MR-invisible objects such as the bed and the MR coils cannot be corrected from MR data. We quantified the effects of a Medres warming bed and a local brain coil on AC for small animal PET/MR imaging using a dedicated Inveon PET tomograph. Additionally, the effect of segmentation-based MRAC of rats in brain regions was evaluated using MR images separately acquired on a preclinical 7T MRI system.

BackgroundPET data-driven attenuation correction (AC) methods, including deep learning, are attractive options for quantitative brain imaging on CT-less brain PET systems and low-dose PET/CT. However, current schemes have performance and practical limitations. We previously developed a CT-less transmission-aided AC that combines coincidences from a weak positron source, and the patient, to estimate attenuation with physics alone. In this work, we aim to optimize and assess this new AC method during human [18F]FDG neuroimaging on whole-body PET/CT.MethodsOur approach, TRansmission-aided μ-map reconstruction (TRU) AC, includes 1) a low-profile and physically fixed transmission source filled with ~ 14 MBq of 18F, 2) a modified maximum likelihood reconstruction of attenuation and activity algorithm, and 3) scatter corrections using the exam data alone. We imaged N = 5 patients with the transmission source, immediately after their clinical [18F]FDG PET/CT. The clinically-consistent protocol included a CT and 10-minute brain-focused PET exam. Using this 10 minutes of patient PET data alone, radiotracer images were reconstructed with the vendor’s algorithm and TRU-AC or CT-AC (reference standard), with all else matched. For quantitative analysis, we placed brain-structure volumes of interest with an atlas, and computed error in mean standardized uptake values of TRU-AC relative to CT-AC.ResultsTRU-AC PET showed qualitatively strong agreement with CT-AC. For the VOI analysis, absolute relative error in standardized uptake values for TRU-AC was within 3.6%, across all brain structures and patients. Normalized root mean square error of activity bias for TRU-AC was 1.8%, and voxel-wise noise in the cerebellum showed a very minor increase of 0.2%. Bland–Altman analysis demonstrated that TRU-AC and CT-AC have statistically significant agreement, assuming a maximum allowed difference of ± 5%.ConclusionsTRU-AC enables quantitative PET for human neuroimaging. This approach may particularly benefit exams where deep learning-based AC schemes show reduced performance, including those focused on radiotracer development, new patient cohorts, and/or pathologies that often lack sufficient training data.Supplementary InformationThe online version contains supplementary material available at 10.1186/s40658-025-00803-x.

MR-based attenuation correction is instrumental for integrated PET/MR imaging. It is generally achieved by segmenting MR images into a set of tissue classes with known attenuation properties (e.g., air, lung, bone, fat, soft tissue). Bone identification with MR imaging is, however, quite challenging, because of the low proton density and fast decay time of bone tissue. The clinical evaluation of a novel, recently published method for zero-echo-time (ZTE)-based MR bone depiction and segmentation in the head is presented here. A new paradigm for MR imaging bone segmentation, based on proton density-weighted ZTE imaging, was disclosed earlier in 2014. In this study, we reviewed the bone maps obtained with this method on 15 clinical datasets acquired with a PET/CT/MR trimodality setup. The CT scans acquired for PET attenuation-correction purposes were used as reference for the evaluation. Quantitative measurements based on the Jaccard distance between ZTE and CT bone masks and qualitative scoring of anatomic accuracy by an experienced radiologist and nuclear medicine physician were performed. The average Jaccard distance between ZTE and CT bone masks evaluated over the entire head was 52% ± 6% (range, 38%-63%). When only the cranium was considered, the distance was 39% ± 4% (range, 32%-49%). These results surpass previously reported attempts with dual-echo ultrashort echo time, for which the Jaccard distance was in the 47%-79% range (parietal and nasal regions, respectively). Anatomically, the calvaria is consistently well segmented, with frequent but isolated voxel misclassifications. Air cavity walls and bone/fluid interfaces with high anatomic detail, such as the inner ear, remain a challenge. This is the first, to our knowledge, clinical evaluation of skull bone identification based on a ZTE sequence. The results suggest that proton density-weighted ZTE imaging is an efficient means of obtaining high-resolution maps of bone tissue with sufficient anatomic accuracy for, for example, PET attenuation correction.

BackgroundA valid photon attenuation correction (AC) method is instrumental for obtaining quantitatively correct PET images. Integrated PET/MR systems provide no direct information on attenuation, and novel methods for MR-based AC (MRAC) are still under investigation. Evaluations of various AC methods have mainly focused on static brain PET acquisitions. In this study, we determined the validity of three MRAC methods in a dynamic PET/MR study of the brain.MethodsNine participants underwent dynamic brain PET/MR scanning using the dopamine transporter radioligand [11C]PE2I. Three MRAC methods were evaluated: single-atlas (Atlas), multi-atlas (MaxProb) and zero-echo-time (ZTE). The 68Ge-transmission data from a previous stand-alone PET scan was used as reference method. Parametric relative delivery (R1) images and binding potential (BPND) maps were generated using cerebellar grey matter as reference region. Evaluation was based on bias in MRAC maps, accuracy and precision of [11C]PE2I BPND and R1 estimates, and [11C]PE2I time-activity curves. BPND was examined for striatal regions and R1 in clusters of regions across the brain.ResultsFor BPND, ZTE-MRAC showed the highest accuracy (bias < 2%) in striatal regions. Atlas-MRAC exhibited a significant bias in caudate nucleus (− 12%) while MaxProb-MRAC revealed a substantial, non-significant bias in the putamen (9%). R1 estimates had a marginal bias for all MRAC methods (− 1.0–3.2%). MaxProb-MRAC showed the largest intersubject variability for both R1 and BPND. Standardized uptake values (SUV) of striatal regions displayed the strongest average bias for ZTE-MRAC (~ 10%), although constant over time and with the smallest intersubject variability. Atlas-MRAC had highest variation in bias over time (+10 to − 10%), followed by MaxProb-MRAC (+5 to − 5%), but MaxProb showed the lowest mean bias. For the cerebellum, MaxProb-MRAC showed the highest variability while bias was constant over time for Atlas- and ZTE-MRAC.ConclusionsBoth Maxprob- and ZTE-MRAC performed better than Atlas-MRAC when using a 68Ge transmission scan as reference method. Overall, ZTE-MRAC showed the highest precision and accuracy in outcome parameters of dynamic [11C]PE2I PET analysis with use of kinetic modelling.

BackgroundIn integrated PET/MR hybrid imaging the evaluation of PET performance characteristics according to the NEMA standard NU 2–2007 is challenging because of incomplete MR-based attenuation correction (AC) for phantom imaging. In this study, a strategy for CT-based AC of the NEMA image quality (IQ) phantom is assessed. The method is systematically evaluated in NEMA IQ phantom measurements on an integrated PET/MR system.MethodsNEMA IQ measurements were performed on the integrated 3.0 Tesla PET/MR hybrid system (Biograph mMR, Siemens Healthcare). AC of the NEMA IQ phantom was realized by an MR-based and by a CT-based method. The suggested CT-based AC uses a template μ-map of the NEMA IQ phantom and a phantom holder for exact repositioning of the phantom on the systems patient table. The PET image quality parameters contrast recovery, background variability, and signal-to-noise ratio (SNR) were determined and compared for both phantom AC methods. Reconstruction parameters of an iterative 3D OP-OSEM reconstruction were optimized for highest lesion SNR in NEMA IQ phantom imaging.ResultsUsing a CT-based NEMA IQ phantom μ-map on the PET/MR system is straightforward and allowed performing accurate NEMA IQ measurements on the hybrid system. MR-based AC was determined to be insufficient for PET quantification in the tested NEMA IQ phantom because only photon attenuation caused by the MR-visible phantom filling but not the phantom housing is considered. Using the suggested CT-based AC, the highest SNR in this phantom experiment for small lesions (<= 13 mm) was obtained with 3 iterations, 21 subsets and 4 mm Gaussian filtering.ConclusionThis study suggests CT-based AC for the NEMA IQ phantom when performing PET NEMA IQ measurements on an integrated PET/MR hybrid system. The superiority of CT-based AC for this phantom is demonstrated by comparison to measurements using MR-based AC. Furthermore, optimized PET image reconstruction parameters are provided for the highest lesion SNR in NEMA IQ phantom measurements.

The aim of this paper is to describe a new automatic method for compensation of metal-implant-induced segmentation errors in MR-based attenuation maps (MRMaps) and to evaluate the quantitative influence of those artifacts on the reconstructed PET activity concentration. The developed method uses a PET-based delineation of the patient contour to compensate metal-implant-caused signal voids in the MR scan that is segmented for PET attenuation correction. PET emission data of 13 patients with metal implants examined in a Philips Ingenuity PET/MR were reconstructed with the vendor-provided method for attenuation correction (MRMaporig, PETorig) and additionally with a method for attenuation correction (MRMapcor, PETcor) developed by our group. MRMaps produced by both methods were visually inspected for segmentation errors. The segmentation errors in MRMaporig were classified into four classes (L1 and L2 artifacts inside the lung and B1 and B2 artifacts inside the remaining body depending on the assigned attenuation coefficients). The average relative SUV differences () between PETorig and PETcor of all regions showing wrong attenuation coefficients in MRMaporig were calculated. Additionally, relative SUVmean differences (εrel) of tracer accumulations in hot focal structures inside or in the vicinity of these regions were evaluated. MRMaporig showed erroneous attenuation coefficients inside the regions affected by metal artifacts and inside the patients' lung in all 13 cases. In MRMapcor, all regions with metal artifacts, except for the sternum, were filled with the soft-tissue attenuation coefficient and the lung was correctly segmented in all patients. MRMapcor only showed small residual segmentation errors in eight patients. (mean ± standard deviation) were: ( − 56 ± 3)% for B1, ( − 43 ± 4)% for B2, (21 ± 18)% for L1, (120 ± 47)% for L2 regions. εrel (mean ± standard deviation) of hot focal structures were: ( − 52 ± 12)% in B1, ( − 45 ± 13)% in B2, (19 ± 19)% in L1, (51 ± 31)% in L2 regions. Consequently, metal-implant-induced artifacts severely disturb MR-based attenuation correction and SUV quantification in PET/MR. The developed algorithm is able to compensate for these artifacts and improves SUV quantification accuracy distinctly.

ObjectiveTo evaluate the frequency of artifacts in MR-based attenuation correction (AC) maps and their impact on the quantitative accuracy of PET-based flow and metabolism measurements in a cohort of consecutive heart failure patients undergoing combined PET/MR imaging. MethodsMyocardial viability studies were performed in 20 patients following a dual-tracer protocol involving the assessment of myocardial perfusion (13N-NH3: 813 ± 86 MBq) and metabolism (18F-FDG: 335 ± 38 MBq). All acquisitions were performed using a fully-integrated PET/MR system, with standard DIXON-attenuation correction (DIXON-AC) mapping for each PET scan. All AC maps were examined for spatial misalignment with the emission data, total lung volume, susceptibility artifacts, and tissue inversion (TI). Misalignment and susceptibility artifacts were corrected using rigid co-registration and retrospective filling of the susceptibility-induced gaps, respectively. The effects of the AC artifacts were evaluated by relative difference measures and perceived changes in clinical interpretations. ResultsAverage respiratory misalignment of (7 ± 4) mm of the PET-emission data and the AC maps was observed in 18 (90%) patients. Substantial changes in the lung volumes of the AC maps were observed in the test–retest analysis (ratio: 1.0 ± 0.2, range: 0.8-1.4). Susceptibility artifacts were observed in 10 (50%) patients, while six (30%) patients had TI artifacts. Average differences of 14 ± 10% were observed for PET images reconstructed with the artifactual AC maps. The combined artifact effects caused false-positive findings in three (15%) patients. ConclusionStandard DIXON-AC maps must be examined carefully for artifacts and misalignment effects prior to AC correction of cardiac PET/MRI studies in order to avoid misinterpretation of biased perfusion and metabolism readings from the PET data.

The goal of this study was to compare MR-based PET patient attenuation correction (AC) to CT-based AC in the head using clinical whole-body FDG-PET patient data obtained from a tri-modality PET/CT & MR setup. The MR-based AC utilizes an atlas-based approach, registering the patient's MR images to a CT-based atlas, producing `pseudoCT' images. Thirteen clinical whole-body FDG patients were included in this study. PET mean activity concentration values were measured and compared in six ~ 15 ml volumes-of-interest throughout the brain tissue. The AC methods compared to CT-based AC were segmentation of the CT (air, fat, soft tissue) and atlas-based MR-AC. Results: PET activity concentration was systematically under-estimated on average by 1.32 kBq/ml (4.9%) when using the segmented CT-based AC, mainly due to lack of attenuation correction for skull bone. Using the atlas-based method, the error was reduced to 0.03 kBq/ml (0.2%) on average. PET image visualization demonstrated spatial variations in activity concentration accuracy induced by the AC methods that were consistent with the approximations in each method. Conclusion: The results demonstrate that the atlas-based AC in the head provides adequate PET quantitation and image quality as compared to methods that do not account for bone.

The goal of this study was to compare MR-based PET patient attenuation correction (AC) to CT-based AC in the head using clinical whole-body FDG-PET patient data obtained from a tri-modality PET/CT & MR setup. The MR-based AC utilizes an atlas-based approach, registering the patient's MR images to a CT-based atlas, producing 'pseudoCT' images. Thirteen clinical whole-body FDG patients were included in this study. PET mean activity concentration values were measured and compared in six ~15ml volumes-of-interest throughout the brain tissue. The AC methods compared to CT-based AC were segmentation of the CT (air, fat, soft tissue) and atlas-based MRAC. Results: PET activity concentration was systematically under-estimated on average by 1.32 kBq/ml (4.9%) when using the segmented CT-based AC, mainly due to lack of attenuation correction for skull bone. Using the atlas-based method, the error was reduced to 0.03 kBq/ml (0.2%) on average. Examination of the resulting PET images shows the spatial variation induced by these methods based upon the approximations made. Conclusion: The results demonstrate that the atlas-based AC in the head provides adequate PET quantitation and image quality as compared to methods which did not include bone.

PurposeThis study evaluates the quantitative effect of improved MR-based attenuation correction (AC), including bone segmentation and the HUGE method for truncation correction in PET/MR whole-body hybrid imaging specifically of oncologic patients with bone metastasis and using various radiotracers.MethodsTwenty-three patients that underwent altogether 28 whole-body PET/MR examinations with findings of bone metastasis were included in this study. Different radiotracers (18F-FDG, 68Ga-PSMA, 68Ga-DOTATOC, 124I–MIBG) were injected according to appropriate clinical indications. Each of the 28 whole-body PET datasets was reconstructed three times using AC with (1) standard four-compartment μ-maps (background air, lung, muscle, and soft tissue), (2) five-compartment μ-maps (adding bone), and (3) six-compartment μ-maps (adding bone and HUGE truncation correction). The SUVmax of each detected bone lesion was measured in each reconstruction to evaluate the quantitative impact of improved MR-based AC. Relative difference images between four- and six-compartment μ-maps were calculated. MR-based HUGE truncation correction was compared with the PET-based MLAA truncation correction method in all patients.ResultsOverall, 69 bone lesions were detected and evaluated. The mean increase in relative difference over all 69 lesions in SUVmax was 5.4 ± 6.4% when comparing the improved six-compartment AC with the standard four-compartment AC. Maximal relative difference of 28.4% was measured in one lesion. Truncation correction with HUGE worked robust and resulted in realistic body contouring in all 28 exams and for all 4 different radiotracers. Truncation correction with MLAA revealed overestimations of arm tissue volume in all PET/MR exams with 18F-FDG radiotracer and failed in all other exams with radiotracers 68Ga-PSMA, 68Ga-DOTATOC, and 124I- MIBG due to limitations in body contour detection.ConclusionImproved MR-based AC, including bone segmentation and HUGE truncation correction in whole-body PET/MR on patients with bone lesions and using various radiotracers, is important to ensure best possible diagnostic image quality and accurate PET quantification. The HUGE method for truncation correction based on MR worked robust and results in realistic body contouring, independent of the radiotracers used.