Predictive uncertainty in deep learning-based MR image reconstruction using deep ensembles: Evaluation on the fastMRI data set.

PURPOSE: To estimate pixel-wise predictive uncertainty for deep learning-based MR image reconstruction and to examine the impact of domain shifts and architecture robustness. METHODS: Uncertainty prediction could provide a measure for robustness of deep learning (DL)-based MR image reconstruction from undersampled data. DL methods bear the risk of inducing reconstruction errors like in-painting of unrealistic structures or missing pathologies. These errors may be obscured by visual realism of DL reconstruction and thus remain undiscovered. Furthermore, most methods are task-agnostic and not well calibrated to domain shifts. We propose a strategy that estimates aleatoric (data) and epistemic (model) uncertainty, which entails training a deep ensemble (epistemic) with nonnegative log-likelihood (aleatoric) loss in addition to the conventional applied losses terms. The proposed procedure can be paired with any DL reconstruction, enabling investigations of their predictive uncertainties on a pixel level. Five different architectures were investigated on the fastMRI database. The impact on the examined uncertainty of in-distributional and out-of-distributional data with changes to undersampling pattern, imaging contrast, imaging orientation, anatomy, and pathology were explored. RESULTS: Predictive uncertainty could be captured and showed good correlation to normalized mean squared error. Uncertainty was primarily focused along the aliased anatomies and on hyperintense and hypointense regions. The proposed uncertainty measure was able to detect disease prevalence shifts. Distinct predictive uncertainty patterns were observed for changing network architectures. CONCLUSION: The proposed approach enables aleatoric and epistemic uncertainty prediction for DL-based MR reconstruction with an interpretable examination on a pixel level.

An Empirical Approach to Derive Water T1 from Multiparametric MR Images Using an Automated Pipeline and Comparison With Liver Stiffness.

BACKGROUND: Water T1 of the liver has been shown to be promising in discriminating the progressive forms of fatty liver diseases, inflammation, and fibrosis, yet proper correction for iron and lipid is required. PURPOSE: To examine the feasibility of an empirical approach for iron and lipid correction when measuring imaging-based T1 and to validate this approach by spectroscopy on in vivo data. STUDY TYPE: Retrospective. POPULATION: Next to mixed lipid-iron phantoms, individuals with different hepatic lipid content were investigated, including people with type 1 diabetes (N = 15, %female = 15.6, age = 43.5 ± 14.0), or type 2 diabetes mellitus (N = 21, %female = 28.9, age = 59.8 ± 9.7) and healthy volunteers (N = 9, %female = 11.1, age = 58.0 ± 8.1). FIELD STRENGTH/SEQUENCES: 3 T, balanced steady-state free precession MOdified Look-Locker Inversion recovery (MOLLI), multi- and dual-echo gradient echo Dixon, gradient echo magnetic resonance elastography (MRE). ASSESSMENT: T1 values were measured in phantoms to determine the respective correction factors. The correction was tested in vivo and validated by proton magnetic resonance spectroscopy (1 H-MRS). The quantification of liver T1 based on automatic segmentation was compared to the T1 values based on manual segmentation. The association of T1 with MRE-derived liver stiffness was evaluated. STATISTICAL TESTS: Bland-Altman plots and intraclass correlation coefficients (ICCs) were used for MOLLI vs. 1 H-MRS agreement and to compare liver T1 values from automatic vs. manual segmentation. Pearson's r correlation coefficients for T1 with hepatic lipids and liver stiffness were determined. A P-value of 0.05 was considered statistically significant. RESULTS: MOLLI T1 values after correction were found in better agreement with the 1 H-MRS-derived water T1 (ICC = 0.60 [0.37; 0.76]) in comparison with the uncorrected T1 values (ICC = 0.18 [-0.09; 0.44]). Automatic quantification yielded similar liver T1 values (ICC = 0.9995 [0.9991; 0.9997]) as with manual segmentation. A significant correlation of T1 with liver stiffness (r = 0.43 [0.11; 0.67]) was found. A marked and significant reduction in the correlation strength of T1 with liver stiffness (r = 0.05 [-0.28; 0.38], P = 0.77) was found after correction for hepatic lipid content. DATA CONCLUSION: Imaging-based correction factors enable accurate estimation of water T1 in vivo. LEVEL OF EVIDENCE: 1 TECHNICAL EFFICACY: Stage 1.

Reducing energy consumption in musculoskeletal MRI using shorter scan protocols, optimized magnet cooling patterns, and deep learning sequences.

OBJECTIVES: The unprecedented surge in energy costs in Europe, coupled with the significant energy consumption of MRI scanners in radiology departments, necessitates exploring strategies to optimize energy usage without compromising efficiency or image quality. This study investigates MR energy consumption and identifies strategies for improving energy efficiency, focusing on musculoskeletal MRI. We assess the potential savings achievable through (1) optimizing protocols, (2) incorporating deep learning (DL) accelerated acquisitions, and (3) optimizing the cooling system. MATERIALS AND METHODS: Energy consumption measurements were performed on two MRI scanners (1.5-T Aera, 1.5-T Sola) in practices in Munich, Germany, between December 2022 and March 2023. Three levels of energy reduction measures were implemented and compared to the baseline. Wilcoxon signed-rank test with Bonferroni correction was conducted to evaluate the impact of sequence scan times and energy consumption. RESULTS: Our findings showed significant energy savings by optimizing protocol settings and implementing DL technologies. Across all body regions, the average reduction in energy consumption was 72% with DL and 31% with economic protocols, accompanied by time reductions of 71% (DL) and 18% (economic protocols) compared to baseline. Optimizing the cooling system during the non-scanning time showed a 30% lower energy consumption. CONCLUSION: Implementing energy-saving strategies, including economic protocols, DL accelerated sequences, and optimized magnet cooling, can significantly reduce energy consumption in MRI scanners. Radiology departments and practices should consider adopting these strategies to improve energy efficiency and reduce costs. CLINICAL RELEVANCE STATEMENT: MRI scanner energy consumption can be substantially reduced by incorporating protocol optimization, DL accelerated acquisition, and optimized magnetic cooling into daily practice, thereby cutting costs and environmental impact. KEY POINTS: Optimization of protocol settings reduced energy consumption by 31% and imaging time by 18%. DL technologies led to a 72% reduction in energy consumption of and a 71% reduction in time, compared to the standard MRI protocol. During non-scanning times, activating Eco power mode (EPM) resulted in a 30% reduction in energy consumption, saving 4881 € ($5287) per scanner annually.

Free-breathing Arterial Spin Labeling MRI for the Detection of Pulmonary Embolism.

Background Arterial spin labeling (ASL) MRI can be used to assess organ perfusion but has yet to be implemented for perfusion evaluation of the lung. Purpose To evaluate pseudo-continuous ASL (PCASL) MRI for the detection of acute pulmonary embolism (PE) and its potential as an alternative to CT pulmonary angiography (CTPA). Materials and Methods Between November 2020 and November 2021, 97 patients (median age, 61 years; 48 women) with suspected PE were enrolled in this prospective study. PCASL MRI was performed within a 72-hour period following CTPA under free-breathing conditions and included three orthogonal planes. The pulmonary trunk was labeled during systole, and the image was acquired during diastole of the subsequent cardiac cycle. Additionally, multisection, coronal, balanced, steady-state free-precession imaging was carried out. Two radiologists blindly assessed overall image quality, artifacts, and diagnostic confidence (five-point Likert scale, 5 = best). Patients were categorized as positive or negative for PE, and a lobe-wise assessment in PCASL MRI and CTPA was conducted. Sensitivity and specificity were calculated on a patient level with the final clinical diagnosis serving as the reference standard. Interchangeability between MRI and CTPA was also tested with use of an individual equivalence index (IEI). Results PCASL MRI was performed successfully in all patients with high scores for image quality, artifact, and diagnostic confidence (κ ≥ .74). Of the 97 patients, 38 were positive for PE. PCASL MRI depicted PE correctly in 35 of 38 patients with three false-positive and three false-negative findings, resulting in a sensitivity of 35 of 38 patients (92% [95% CI: 79, 98]) and a specificity of 56 of 59 patients (95% [95% CI: 86, 99]). Interchangeability analysis revealed an IEI of 2.6% (95% CI: 1.2, 3.8). Conclusion Free-breathing pseudo-continuous arterial spin labeling MRI depicted abnormal lung perfusion caused by acute pulmonary embolism and may be useful as a contrast material-free alternative to CT pulmonary angiography for selected patients. German Clinical Trials Register no. DRKS00023599 © RSNA, 2023.

Identification of radiomic biomarkers in a set of four skeletal muscle groups on Dixon MRI of the NAKO MR study.

In this work, we propose a processing pipeline for the extraction and identification of meaningful radiomics biomarkers in skeletal muscle tissue as displayed using Dixon-weighted MRI. Diverse and robust radiomics features can be identified that may be of aid in the accurate quantification e.g. varying degrees of sarcopenia in respective muscles of large cohorts. As such, the approach comprises the texture feature extraction from raw data based on well established approaches, such as a nnU-Net neural network and the Pyradiomics toolbox, a subsequent selection according to adequate conditions for the muscle tissue of the general population, and an importance-based ranking to further narrow the amount of meaningful features with respect to auxiliary targets. The performance was investigated with respect to the included auxiliary targets, namely age, body mass index (BMI), and fat fraction (FF). Four skeletal muscles with different fiber architecture were included: the mm. glutaei, m. psoas, as well as the extensors and adductors of the thigh. The selection allowed for a reduction from 1015 available texture features to 65 for age, 53 for BMI, and 36 for FF from the available fat/water contrast images considering all muscles jointly. Further, the dependence of the importance rankings calculated for the auxiliary targets on validation sets (in a cross-validation scheme) was investigated by boxplots. In addition, significant differences between subgroups of respective auxiliary targets as well as between both sexes were shown to be present within the ten lowest ranked features by means of Kruskal-Wallis H-tests and Mann-Whitney U-tests. The prediction performance for the selected features and the ranking scheme were verified on validation sets by a random forest based multi-class classification, with strong area under the curve (AUC) values of the receiver operator characteristic (ROC) of 73.03 ± 0.70 % and 73.63 ± 0.70 % for the water and fat images in age, 80.68 ± 0.30 % and 88.03 ± 0.89 % in BMI, as well as 98.36 ± 0.03 % and 98.52 ± 0.09 % in FF.

Physics-Driven Deep Learning for Computational Magnetic Resonance Imaging: Combining physics and machine learning for improved medical imaging.

Physics-driven deep learning methods have emerged as a powerful tool for computational magnetic resonance imaging (MRI) problems, pushing reconstruction performance to new limits. This article provides an overview of the recent developments in incorporating physics information into learning-based MRI reconstruction. We consider inverse problems with both linear and non-linear forward models for computational MRI, and review the classical approaches for solving these. We then focus on physics-driven deep learning approaches, covering physics-driven loss functions, plug-and-play methods, generative models, and unrolled networks. We highlight domain-specific challenges such as real- and complex-valued building blocks of neural networks, and translational applications in MRI with linear and non-linear forward models. Finally, we discuss common issues and open challenges, and draw connections to the importance of physics-driven learning when combined with other downstream tasks in the medical imaging pipeline.

Deep-learning based super-resolution for 3D isotropic coronary MR angiography in less than a minute.

PURPOSE: To develop and evaluate a novel and generalizable super-resolution (SR) deep-learning framework for motion-compensated isotropic 3D coronary MR angiography (CMRA), which allows free-breathing acquisitions in less than a minute. METHODS: Undersampled motion-corrected reconstructions have enabled free-breathing isotropic 3D CMRA in ~5-10 min acquisition times. In this work, we propose a deep-learning-based SR framework, combined with non-rigid respiratory motion compensation, to shorten the acquisition time to less than 1 min. A generative adversarial network (GAN) is proposed consisting of two cascaded Enhanced Deep Residual Network generator, a trainable discriminator, and a perceptual loss network. A 16-fold increase in spatial resolution is achieved by reconstructing a high-resolution (HR) isotropic CMRA (0.9 mm3 or 1.2 mm3 ) from a low-resolution (LR) anisotropic CMRA (0.9 × 3.6 × 3.6 mm3 or 1.2 × 4.8 × 4.8 mm3 ). The impact and generalization of the proposed SRGAN approach to different input resolutions and operation on image and patch-level is investigated. SRGAN was evaluated on a retrospective downsampled cohort of 50 patients and on 16 prospective patients that were scanned with LR-CMRA in ~50 s under free-breathing. Vessel sharpness and length of the coronary arteries from the SR-CMRA is compared against the HR-CMRA. RESULTS: SR-CMRA showed statistically significant (P < .001) improved vessel sharpness 34.1% ± 12.3% and length 41.5% ± 8.1% compared with LR-CMRA. Good generalization to input resolution and image/patch-level processing was found. SR-CMRA enabled recovery of coronary stenosis similar to HR-CMRA with comparable qualitative performance. CONCLUSION: The proposed SR-CMRA provides a 16-fold increase in spatial resolution with comparable image quality to HR-CMRA while reducing the predictable scan time to <1 min.

Complementary time-frequency domain networks for dynamic parallel MR image reconstruction.

PURPOSE: To introduce a novel deep learning-based approach for fast and high-quality dynamic multicoil MR reconstruction by learning a complementary time-frequency domain network that exploits spatiotemporal correlations simultaneously from complementary domains. THEORY AND METHODS: Dynamic parallel MR image reconstruction is formulated as a multivariable minimization problem, where the data are regularized in combined temporal Fourier and spatial (x-f) domain as well as in spatiotemporal image (x-t) domain. An iterative algorithm based on variable splitting technique is derived, which alternates among signal de-aliasing steps in x-f and x-t spaces, a closed-form point-wise data consistency step and a weighted coupling step. The iterative model is embedded into a deep recurrent neural network which learns to recover the image via exploiting spatiotemporal redundancies in complementary domains. RESULTS: Experiments were performed on two datasets of highly undersampled multicoil short-axis cardiac cine MRI scans. Results demonstrate that our proposed method outperforms the current state-of-the-art approaches both quantitatively and qualitatively. The proposed model can also generalize well to data acquired from a different scanner and data with pathologies that were not seen in the training set. CONCLUSION: The work shows the benefit of reconstructing dynamic parallel MRI in complementary time-frequency domains with deep neural networks. The method can effectively and robustly reconstruct high-quality images from highly undersampled dynamic multicoil data ( 16× and 24× yielding 15 s and 10 s scan times respectively) with fast reconstruction speed (2.8 seconds). This could potentially facilitate achieving fast single-breath-hold clinical 2D cardiac cine imaging.

Fully self-gated free-running 3D Cartesian cardiac CINE with isotropic whole-heart coverage in less than 2 min.

PURPOSE: To develop a novel fast water-selective free-breathing 3D Cartesian cardiac CINE scan with full self-navigation and isotropic whole-heart (WH) coverage. METHODS: A free-breathing 3D Cartesian cardiac CINE scan with a water-selective balanced steady-state free precession and a continuous (non-ECG-gated) variable-density Cartesian sampling with spiral profile ordering, out-inward sampling and acquisition-adaptive alternating tiny golden and golden angle increment between spiral arms is proposed. Data is retrospectively binned based on respiratory and cardiac self-navigation signals. A translational respiratory-motion-corrected and cardiac-motion-resolved image is reconstructed with a multi-bin patch-based low-rank reconstruction (MB-PROST) within about 15 min. A respiratory-motion-resolved approach is also investigated. The proposed 3D Cartesian cardiac CINE is acquired in sagittal orientation in 1 min 50 s for 1.9 mm3 isotropic WH coverage. Left ventricular (LV) function parameters and image quality derived from a blinded reading of the proposed 3D CINE framework are compared against conventional multi-slice 2D CINE imaging in 10 healthy subjects and 10 patients with suspected cardiovascular disease. RESULTS: The proposed framework provides free-breathing 3D cardiac CINE images with 1.9 mm3 spatial and about 45 ms temporal resolution in a short acquisition time (<2 min). LV function parameters derived from 3D CINE were in good agreement with 2D CINE (10 healthy subjects and 10 patients). Bias and confidence intervals were obtained for end-systolic volume, end-diastolic volume and ejection fraction of 0.1 ± 3.5 mL, -0.6 ± 8.2 mL and -0.1 ± 2.2%, respectively. CONCLUSION: The proposed framework enables isotropic 3D Cartesian cardiac CINE under free breathing for fast assessment of cardiac anatomy and function.

Isotropic 3D Cartesian single breath-hold CINE MRI with multi-bin patch-based low-rank reconstruction.

PURPOSE: To develop a novel acquisition and reconstruction framework for isotropic 3D Cartesian cardiac CINE within a single breath-hold for left ventricle (LV) and whole-heart coverage. METHODS: A variable-density Cartesian acquisition with spiral profile ordering, out-inward sampling, and acquisition-adaptive alternating tiny golden/golden angle increment between spiral arms is proposed to provide incoherent and nonredundant sampling within and among cardiac phases. A novel multi-bin patch-based low-rank reconstruction, named MB-PROST, is proposed to exploit redundant information on a local (within a patch), nonlocal (similar patches within a spatial neighborhood), and temporal (among all cardiac phases) scale with an implicit motion alignment among patches. The proposed multi-bin patch-based low-rank reconstruction reconstruction is compared against compressed sensing reconstruction, whereas LV function parameters derived from the proposed 3D CINE framework are compared against those estimated from conventional multislice 2D CINE imaging in 10 healthy subjects and 15 patients. RESULTS: The proposed framework provides 3D cardiac CINE images with high spatial (1.9 mm3 ) and temporal resolution (˜50 ms) in a single breath-hold of Ëœ20 s for LV and Ëœ26 s for whole-heart coverage in healthy subjects. Shorter breath-hold durations of Ëœ13 to 15 s are feasible for LV coverage with slightly anisotropic resolution (1.9 × 1.9 × 2.5 mm) in patients. LV function parameters derived from 3D CINE were in good agreement with 2D CINE, with a bias of -0.1 mL/0.1 mL, -0.9 mL/-1.0 mL, -0.1%/-0.8%; and confidence intervals of ±1.7 mL/±3.7 mL, ±1.2 mL/±2.6 mL, and ±1.2%/±3.6% (10 healthy subjects/15 patients) for end-systolic volume, end-diastolic volume, and ejection fraction, respectively. CONCLUSION: The proposed framework enables 3D isotropic cardiac CINE in a single breath-hold scan of Ëœ20 s/˜26 s for LV/whole-heart coverage, showing good agreement with clinical 2D CINE scans in terms of LV functional assessment.

Imaging Pulmonary Blood Flow Using Pseudocontinuous Arterial Spin Labeling (PCASL) With Balanced Steady-State Free-Precession (bSSFP) Readout at 1.5T.

BACKGROUND: Quantitative assessment of pulmonary blood flow and visualization of its temporal and spatial distribution without contrast media is of clinical significance. PURPOSE: To assess the potential of electrocardiogram (ECG)-triggered pseudocontinuous arterial spin labeling (PCASL) imaging with balanced steady-state free-precession (bSSFP) readout to measure lung perfusion under free-breathing (FB) conditions and to study temporal and spatial characteristics of pulmonary blood flow. STUDY TYPE: Prospective, observational. SUBJECTS: Fourteen volunteers; three patients with pulmonary embolism. FIELD STRENGTH/SEQUENCES: 1.5T, PCASL-bSSFP. ASSESSMENT: The pulmonary trunk was labeled during systole. The following examinations were performed: 1) FB and timed breath-hold (TBH) examinations with a postlabeling delay (PLD) of 1000 msec, and 2) TBH examinations with multiple PLDs (100-1500 msec). Scan-rescan measurements were performed in four volunteers and one patient. Images were registered and the perfusion was evaluated in large vessels, small vessels, and parenchyma. Mean structural similarity indices (MSSIM) was computed and time-to-peak (TTP) of parenchymal perfusion in multiple PLDs was evaluated. Image quality reading was performed with three independent blinded readers. STATISTICAL TESTS: Wilcoxon test to compare MSSIM, perfusion, and Likert scores. Spearman's correlation to correlate TTP and cardiac cycle duration. The repeatability coefficient (RC) and within-subject coefficient of variation (wCV) for scan-rescan measurements. Intraclass correlation coefficient (ICC) for interreader agreement. RESULTS: Image registration resulted in a significant (P < 0.05) increase of MSSIM. FB perfusion values were 6% higher than TBH (3.28 ± 1.09 vs. 3.10 ± 0.99 mL/min/mL). TTP was highly correlated with individuals' cardiac cycle duration (Spearman = 0.89, P < 0.001). RC and wCV were better for TBH than FB (0.13-0.19 vs. 0.47-1.54 mL/min/mL; 6-7 vs. 19-60%). Image quality was rated very good, with ICCs 0.71-0.89. DATA CONCLUSION: ECG-triggered PCASL-bSSFP imaging of the lung at 1.5T can provide very good image quality and quantitative perfusion maps even under FB. The course of labeled blood through the lung shows a strong dependence on the individuals' cardiac cycle duration. LEVEL OF EVIDENCE: 2 TECHNICAL EFFICACY STAGE: 2 J. MAGN. RESON. IMAGING 2020;52:1767-1782.

3D Cartesian fast interrupted steady-state (FISS) imaging.

PURPOSE: To enable intrinsic and efficient fat suppression in 3D Cartesian fast interrupted steady-state (FISS) acquisitions. METHODS: A periodic interruption of the balanced steady-state free precession (bSSFP) readout train (FISS) has been previously proposed for 2D radial imaging. FISS modulates the bSSFP frequency response pattern in terms of shape, width and location of stop band (attenuated transverse magnetization). Depending on the FISS interruption rate, the stop band characteristic can be exploited to suppress the fat spectrum at 3.5 ppm, thus yielding intrinsic fat suppression. For conventional 2D Cartesian sampling, ghosting/aliasing artifacts along phase-encoding direction have been reported. In this work, we propose to extend FISS to 3D Cartesian imaging and report countermeasures for the previously observed ghosting/aliasing artifacts. Key parameters (dummy prepulses, spatial resolution, and interruption rate) are investigated to optimize fat suppression and image quality. FISS behavior is examined using extended phase graph simulations to recommend parametrizations which are validated in phantom and in vivo measurements on a 1.5T MRI scanner for 3 applications: upper thigh angiography, abdominal imaging, and free-running 5D CINE. RESULTS: Using optimized parameters, 3D Cartesian FISS provides homogeneous and consistent fat suppression for all 3 applications. In upper thigh angiography, vessel structures can be recovered in FISS that are obscured in bSSFP. Fat suppression in free-running cardiac CINE resulted in less fat-related motion aliasing and yielded better image quality. CONCLUSION: 3D Cartesian FISS is feasible and offers homogeneous intrinsic fat suppression for selected imaging parameters without the need for dedicated preparation pulses, making it a promising candidate for free-running fat-suppressed imaging.

Retrospective correction of motion-affected MR images using deep learning frameworks.

PURPOSE: Motion is 1 extrinsic source for imaging artifacts in MRI that can strongly deteriorate image quality and, thus, impair diagnostic accuracy. In addition to involuntary physiological motion such as respiration and cardiac motion, intended and accidental patient movements can occur. Any impairment by motion artifacts can reduce the reliability and precision of the diagnosis and a motion-free reacquisition can become time- and cost-intensive. Numerous motion correction strategies have been proposed to reduce or prevent motion artifacts. These methods have in common that they need to be applied during the actual measurement procedure with a-priori knowledge about the expected motion type and appearance. For retrospective motion correction and without the existence of any a-priori knowledge, this problem is still challenging. METHODS: We propose the use of deep learning frameworks to perform retrospective motion correction in a reference-free setting by learning from pairs of motion-free and motion-affected images. For this image-to-image translation problem, we propose and compare a variational auto encoder and generative adversarial network. Feasibility and influences of motion type and optimal architecture are investigated by blinded subjective image quality assessment and by quantitative image similarity metrics. RESULTS: We observed that generative adversarial network-based motion correction is feasible producing near-realistic motion-free images as confirmed by blinded subjective image quality assessment. Generative adversarial network-based motion correction accordingly resulted in images with high evaluation metrics (normalized root mean squared error <0.08, structural similarity index >0.8, normalized mutual information >0.9). CONCLUSION: Deep learning-based retrospective restoration of motion artifacts is feasible resulting in near-realistic motion-free images. However, the image translation task can alter or hide anatomical features and, therefore, the clinical applicability of this technique has to be evaluated in future studies.

Independent brain 18F-FDG PET attenuation correction using a deep learning approach with Generative Adversarial Networks.

OBJECTIVE: Attenuation correction (AC) of positron emission tomography (PET) data poses a challenge when no transmission data or computed tomography (CT) data are available, e.g. in stand alone PET scanners or PET/magnetic resonance imaging (MRI). In these cases, external imaging data or morphological imaging data are normally used for the generation of attenuation maps. Newly introduced machine learning methods however may allow for direct estimation of attenuation maps from non attenuation-corrected PET data (PETNAC). Our purpose was thus to establish and evaluate a method for independent AC of brain fluorine-18-fluorodeoxyglucose (18F-FDG) PET images only based on PETNAC using Generative Adversarial Networks (GAN). SUBJECTS AND METHODS: After training of the deep learning GAN framework on a paired training dataset of PETNAC and the corresponding CT images of the head from 50 patients, pseudo-CT images were generated from PETNAC of 40 validation patients, of which 20 were used for technical validation and 20 stemming from patients with CNS disorders were used for clinical validation. Pseudo-CT was used for subsequent AC of these validation data sets resulting in independently attenuation-corrected PET data. RESULTS: Visual inspection revealed a high degree of resemblance of generated pseudo-CT images compared to the acquired CT images in all validation data sets, with minor differences in individual anatomical details. Quantitative analyses revealed minimal underestimation below 5% of standardized uptake value (SUV) in all brain regions in independently attenuation-corrected PET data compared to the reference PET images. Color-coded error maps showed no regional bias and only minimal average errors around ±0%. Using independently attenuation-corrected PET data, no differences in image-based diagnoses were observed in 20 patients with neurological disorders compared to the reference PET images. CONCLUSION: Independent AC of brain 18F-FDG PET is feasible with high accuracy using the proposed, easy to implement deep learning framework. Further evaluation in clinical cohorts will be necessary to assess the clinical performance of this method.

Automated reference-free detection of motion artifacts in magnetic resonance images.

OBJECTIVES: Our objectives were to provide an automated method for spatially resolved detection and quantification of motion artifacts in MR images of the head and abdomen as well as a quality control of the trained architecture. MATERIALS AND METHODS: T1-weighted MR images of the head and the upper abdomen were acquired in 16 healthy volunteers under rest and under motion. Images were divided into overlapping patches of different sizes achieving spatial separation. Using these patches as input data, a convolutional neural network (CNN) was trained to derive probability maps for the presence of motion artifacts. A deep visualization offers a human-interpretable quality control of the trained CNN. Results were visually assessed on probability maps and as classification accuracy on a per-patch, per-slice and per-volunteer basis. RESULTS: On visual assessment, a clear difference of probability maps was observed between data sets with and without motion. The overall accuracy of motion detection on a per-patch/per-volunteer basis reached 97%/100% in the head and 75%/100% in the abdomen, respectively. CONCLUSION: Automated detection of motion artifacts in MRI is feasible with good accuracy in the head and abdomen. The proposed method provides quantification and localization of artifacts as well as a visualization of the learned content. It may be extended to other anatomic areas and used for quality assurance of MR images.

Self-navigated 4D cartesian imaging of periodic motion in the body trunk using partial k-space compressed sensing.

PURPOSE: To enable fast and flexible high-resolution four-dimensional (4D) MRI of periodic thoracic/abdominal motion for motion visualization or motion-corrected imaging. METHODS: We proposed a Cartesian three-dimensional k-space sampling scheme that acquires a random combination of k-space lines in the ky/kz plane. A partial Fourier-like constraint compacts the sampling space to one half of k-space. The central k-space line is periodically acquired to allow an extraction of a self-navigated respiration signal used to populate a k-space of multiple breathing positions. The randomness of the acquisition (induced by periodic breathing pattern) yields a subsampled k-space that is reconstructed using compressed sensing. Local image evaluations (coefficient of variation and slope steepness through organs) reveal information about motion resolvability. Image quality is inspected by a blinded reading. Sequence and reconstruction method are made publicly available. RESULTS: The method is able to capture and reconstruct 4D images with high image quality and motion resolution within a short scan time of less than 2 min. These findings are supported by restricted-isometry-property analysis, local image evaluation, and blinded reading. CONCLUSION: The proposed method provides a clinical feasible setup to capture periodic respiratory motion with a fast acquisition protocol and can be extended by further surrogate signals to capture additional periodic motions. Retrospective parametrization allows for flexible tuning toward the targeted applications. Magn Reson Med 78:632-644, 2017. © 2016 International Society for Magnetic Resonance in Medicine.

Combined unsupervised-supervised classification of multiparametric PET/MRI data: application to prostate cancer.

Multiparametric medical imaging data can be large and are often complex. Machine learning algorithms can assist in image interpretation when reliable training data exist. In most cases, however, knowledge about ground truth (e.g. histology) and thus training data is limited, which makes application of machine learning algorithms difficult. The purpose of this study was to design and implement a learning algorithm for classification of multidimensional medical imaging data that is robust and accurate even with limited prior knowledge and that allows for generalization and application to unseen data. Local prostate cancer was chosen as a model for application and validation. 16 patients underwent combined simultaneous [(11) C]-choline positron emission tomography (PET)/MRI. The following imaging parameters were acquired: T2 signal intensities, apparent diffusion coefficients, parameters Ktrans and Kep from dynamic contrast-enhanced MRI, and PET standardized uptake values (SUVs). A spatially constrained fuzzy c-means algorithm (sFCM) was applied to the single datasets and the resulting labeled data were used for training of a support vector machine (SVM) classifier. Accuracy and false positive and false negative rates of the proposed algorithm were determined in comparison with manual tumor delineation. For five of the 16 patients rates were also determined in comparison with the histopathological standard of reference. The combined sFCM/SVM algorithm proposed in this study revealed reliable classification results consistent with the histopathological reference standard and comparable to those of manual tumor delineation. sFCM/SVM generally performed better than unsupervised sFCM alone. We observed an improvement in accuracy with increasing number of imaging parameters used for clustering and SVM training. In particular, including PET SUVs as an additional parameter markedly improved classification results. A variety of applications are conceivable, especially for imaging of tissues without easily available histopathological correlation.

Motion-Compensated MR CINE Reconstruction With Reconstruction-Driven Motion Estimation.

In cardiac CINE, motion-compensated MR reconstruction (MCMR) is an effective approach to address highly undersampled acquisitions by incorporating motion information between frames. In this work, we propose a novel perspective for addressing the MCMR problem and a more integrated and efficient solution to the MCMR field. Contrary to state-of-the-art (SOTA) MCMR methods which break the original problem into two sub-optimization problems, i.e. motion estimation and reconstruction, we formulate this problem as a single entity with one single optimization. Our approach is unique in that the motion estimation is directly driven by the ultimate goal, reconstruction, but not by the canonical motion-warping loss (similarity measurement between motion-warped images and target images). We align the objectives of motion estimation and reconstruction, eliminating the drawbacks of artifacts-affected motion estimation and therefore error-propagated reconstruction. Further, we can deliver high-quality reconstruction and realistic motion without applying any regularization/smoothness loss terms, circumventing the non-trivial weighting factor tuning. We evaluate our method on two datasets: 1) an in-house acquired 2D CINE dataset for the retrospective study and 2) the public OCMR cardiac dataset for the prospective study. The conducted experiments indicate that the proposed MCMR framework can deliver artifact-free motion estimation and high-quality MR images even for imaging accelerations up to 20x, outperforming SOTA non-MCMR and MCMR methods in both qualitative and quantitative evaluation across all experiments. The code is available at https://github.com/JZPeterPan/MCMR-Recon-Driven-Motion.