Image-to-image translation is considered a new frontier in the field of medical image analysis, with numerous potential applications. However, a large portion of recent approaches offers individualized solutions based on specialized task-specific architectures or require refinement through non-end-to-end training. In this paper, we propose a new framework, named MedGAN, for medical image-to-image translation which operates on the image level in an end-toend manner. MedGAN builds upon recent advances in the field of generative adversarial networks (GANs) by merging the adversarial framework with a new combination of non-adversarial losses. We utilize a discriminator network as a trainable feature extractor which penalizes the discrepancy between the translated medical images and the desired modalities. Moreover, style-transfer losses are utilized to match the textures and fine-structures of the desired target images to the translated images. Additionally, we present a new generator architecture, titled CasNet, which enhances the sharpness of the translated medical outputs through progressive refinement via encoder-decoder pairs. Without any application-specific modifications, we apply MedGAN on three different tasks: PET-CT translation, correction of MR motion artefacts and PET image denoising. Perceptual analysis by radiologists and quantitative evaluations illustrate that the MedGAN outperforms other existing translation approaches.

cardiac cine magnetic resonance imaging is the gold-standard for the assessment of cardiac function. Imaging accelerations have shown to enable 3D CINE with left ventricular (LV) coverage in a single breath-hold. However, 3D imaging remains limited to anisotropic resolution and long reconstruction times. Recently deep learning has shown promising results for computationally efficient reconstructions of highly accelerated 2D CINE imaging. In this work, we propose a novel 4D (3D + time) deep learning-based reconstruction network, termed 4D CINENet, for prospectively undersampled 3D Cartesian CINE imaging. CINENet is based on (3 + 1)D complex-valued spatio-temporal convolutions and multi-coil data processing. We trained and evaluated the proposed cinenet on in-house acquired 3D CINE data of 20 healthy subjects and 15 patients with suspected cardiovascular disease. The proposed cinenet network outperforms iterative reconstructions in visual image quality and contrast (+ 67% improvement). We found good agreement in LV function (bias ± 95% confidence) in terms of end-systolic volume (0 ± 3.3 ml), end-diastolic volume (− 0.4 ± 2.0 ml) and ejection fraction (0.1 ± 3.2%) compared to clinical gold-standard 2D CINE, enabling single breath-hold isotropic 3D CINE in less than 10 s scan and ~ 5 s reconstruction time. Cardiac CINE magnetic resonance imaging (MRI) is the gold standard for the assessment of cardiac morphology and function. Conventionally, multi-slice 2D CINE imaging is performed under multiple breath-holds to achieve left ventricular (LV) coverage. For fast LV coverage only a few (~ 12) short-axis 2D slices with anisotropic resolution in the slice direction are acquired throughout multiple breath-holds of < 15 s duration each. Imperfect (e.g. drifts) or varying breath-hold positions and the anisotropic image resolution can cause slice misalignments which may lead to staircasing artifacts and erroneous assessment of the ventricular volume. The LV function assessment is assessed by epicardial and endocardial segmentation of the images in short-axis orientation. Indeed, the anisotropic resolution of the short-axis 2D CINE does not allow for reformats to arbitrary orientations. Further images in other long axis orientations are required for a comprehensive assessment of cardiac morphology and function which in turn requires multiple acquisitions to be performed in several geometric views and thereby increasing overall planning and scan time. To overcome these limitations, 2D 1,2 and 3D 3-5 free-breathing cardiac CINE imaging with retrospective motion correction have been proposed to minimize slice misalignment and improve patient comfort. Data are acquired under free-breathing and respiratory and cardiac motion is resolved retrospectively which comes however at the expense of a prolonged scan time in the order of several minutes. Moreover, these approaches usually require long reconstruction times associated with the high-dimensional (spatial, respiratory temporal and cardiac temporal) data processing or with the nat...

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.

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.

Background: Cardiovascular magnetic resonance (CMR) T1ρ mapping can be used to detect ischemic or nonischemic cardiomyopathy without the need of exogenous contrast agents. Current 2D myocardial T1ρ mapping requires multiple breath-holds and provides limited coverage. Respiratory gating by diaphragmatic navigation has recently been exploited to enable free-breathing 3D T1ρ mapping, which, however, has low acquisition efficiency and may result in unpredictable and long scan times. This study aims to develop a fast respiratory motioncompensated 3D whole-heart myocardial T1ρ mapping technique with high spatial resolution and predictable scan time. Methods: The proposed electrocardiogram (ECG)-triggered T1ρ mapping sequence is performed under freebreathing using an undersampled variable-density 3D Cartesian sampling with spiral-like order. Preparation pulses with different T1ρ spin-lock times are employed to acquire multiple T1ρ-weighted images. A saturation prepulse is played at the start of each heartbeat to reset the magnetization before T1ρ preparation. Image navigators are employed to enable beat-to-beat 2D translational respiratory motion correction of the heart for each T1ρ-weighted dataset, after which, 3D translational registration is performed to align all T1ρ-weighted volumes. Undersampled reconstruction is performed using a multi-contrast 3D patch-based low-rank algorithm. The accuracy of the proposed technique was tested in phantoms and in vivo in 11 healthy subjects in comparison with 2D T1ρ mapping. The feasibility of the proposed technique was further investigated in 3 patients with suspected cardiovascular disease. Breath-hold late-gadolinium enhanced (LGE) images were acquired in patients as reference for scar detection. Results: Phantoms results revealed that the proposed technique provided accurate T1ρ values over a wide range of simulated heart rates in comparison to a 2D T1ρ mapping reference. Homogeneous 3D T1ρ maps were obtained for healthy subjects, with septal T1ρ of 58.0 ± 4.1 ms which was comparable to 2D breath-hold measurements (57.6 ± 4.7 ms, P = 0.83). Myocardial scar was detected in 1 of the 3 patients, and increased T1ρ values (87.4 ± 5.7 ms) were observed in the infarcted region. Conclusions: An accelerated free-breathing 3D whole-heart T1ρ mapping technique was developed with high respiratory scan efficiency and near-isotropic spatial resolution (1.7 × 1.7 × 2 mm 3) in a clinically feasible scan time of~6 mins. Preliminary patient results suggest that the proposed technique may find applications in non-contrast myocardial tissue characterization.

Image-to-image translation is a new field in computer vision with multiple potential applications in the medical domain. However, for supervised image translation frameworks, co-registered datasets, paired in a pixel-wise sense, are required. This is often difficult to acquire in realistic medical scenarios. On the other hand, unsupervised translation frameworks often result in blurred translated images with unrealistic details. In this work, we propose a new unsupervised translation framework which is titled Cycle-MedGAN. The proposed framework utilizes new non-adversarial cycle losses which direct the framework to minimize the textural and perceptual discrepancies in the translated images. Qualitative and quantitative comparisons against other unsupervised translation approaches demonstrate the performance of the proposed framework for PET-CT translation and MR motion correction.

No abstract

Background: Attenuation correction (AC) of PET data is usually performed using a second imaging for the generation of attenuation maps. In certain situations however-when CT-or MR-derived attenuation maps are corrupted or CT acquisition solely for the purpose of AC shall be avoided-it would be of value to have the possibility of obtaining attenuation maps only based on PET information. The purpose of this study was to thus develop, implement, and evaluate a deep learning-based method for whole body [ 18 F]FDG-PET AC which is independent of other imaging modalities for acquiring the attenuation map. Methods: The proposed method is investigated on whole body [ 18 F]FDG-PET data using a Generative Adversarial Networks (GAN) deep learning framework. It is trained to generate pseudo CT images (CT GAN) based on paired training data of non-attenuation corrected PET data (PET NAC) and corresponding CT data. Generated pseudo CTs are then used for subsequent PET AC. One hundred data sets of whole body PET NAC and corresponding CT were used for training. Twenty-five PET/CT examinations were used as test data sets (not included in training). On these test data sets, AC of PET was performed using the acquired CT as well as CT GAN resulting in the corresponding PET data sets PET AC and PET GAN. CT GAN and PET GAN were evaluated qualitatively by visual inspection and by visual analysis of color-coded difference maps. Quantitative analysis was performed by comparison of organ and lesion SUVs between PET AC and PET GAN. Results: Qualitative analysis revealed no major SUV deviations on PET GAN for most anatomic regions; visually detectable deviations were mainly observed along the diaphragm and the lung border. Quantitative analysis revealed mean percent deviations of SUVs on PET GAN of − 0.8 ± 8.6% over all organs (range [− 30.7%, + 27.1%]). Mean lesion SUVs showed a mean deviation of 0.9 ± 9.2% (range [− 19.6%, + 29.2%]). Conclusion: Independent AC of whole body [ 18 F]FDG-PET is feasible using the proposed deep learning approach yielding satisfactory PET quantification accuracy. Further clinical validation is necessary prior to implementation in clinical routine applications.