Scattered radiation is a major concern impacting X-ray image-guided procedures in two ways. First, back-scatter significantly contributes to patient (skin) dose during complicated interventions. Second, forward-scattered radiation reduces contrast in projection images and introduces artifacts in 3-D reconstructions. While conventionally employed anti-scatter grids improve image quality by blocking X-rays, the additional attenuation due to the anti-scatter grid at the detector needs to be compensated for by a higher patient entrance dose. This also increases the room dose affecting the staff caring for the patient. For skin dose quantification, back-scatter is usually accounted for by applying pre-determined scalar back-scatter factors or linear point spread functions to a primary kerma forward projection onto a patient surface point. However, as patients come in different shapes, the generalization of conventional methods is limited. Here, we propose a novel approach combining conventional techniques with learning-based methods to simultaneously estimate the forward-scatter reaching the detector as well as the back-scatter affecting the patient skin dose. Knowing the forward-scatter, we can correct X-ray projections, while a good estimate of the back-scatter component facilitates an improved skin dose assessment. To simultaneously estimate forward-scatter as well as back-scatter, we propose a multi-task approach for joint back- and forward-scatter estimation by combining X-ray physics with neural networks. We show that, in theory, highly accurate scatter estimation in both cases is possible. In addition, we identify research directions for our multi-task framework and learning-based scatter estimation in general.

Continuous protocols for cardiac magnetic resonance imaging enable sampling of the cardiac anatomy simultaneously resolved into cardiac phases. To avoid respiration artifacts, associated motion during the scan has to be compensated for during reconstruction. In this paper, we propose a sampling adaption to acquire 2D respiration information during a continuous scan. Further, we develop a pipeline to extract the different respiration states from the acquired signals, which are used to reconstruct data from one respiration phase. Our results show the benefit of the proposed workflow on the image quality compared to no respiration compensation, as well as a previous 1D respiration navigation approach.

X-ray scatter compensation is a very desirable technique in flat-panel X-ray imaging and cone-beam computed tomography. State-of-the-art U-net based scatter removal approaches yielded promising results. However, as there are no physics' constraints applied to the output of the U-Net, it cannot be ruled out that it yields spurious results. Unfortunately, in the context of medical imaging, those may be misleading and could lead to wrong conclusions. To overcome this problem, we propose to embed B-splines as a known operator into neural networks. This inherently constrains their predictions to well-behaved and smooth functions. In a study using synthetic head and thorax data as well as real thorax phantom data, we found that our approach performed on par with U-net when comparing both algorithms based on quantitative performance metrics. However, our approach not only reduces runtime and parameter complexity, but we also found it much more robust to unseen noise levels. While the U-net responded with visible artifacts, the proposed approach preserved the X-ray signal's frequency characteristics.

For the clinical assessment of cardiac vitality, time-continuous tomographic imaging of the heart is used. To further detect e.g., pathological tissue, multiple imaging contrasts enable a thorough diagnosis using magnetic resonance imaging (MRI). For this purpose, time-continous and multi-contrast imaging protocols were proposed. The acquired signals are binned using navigation approaches for a motion-resolved reconstruction. Mostly, external sensors such as electrocardiograms (ECG) are used for navigation, leading to additional workflow efforts. Recent sensor-free approaches are based on pipelines requiring prior knowledge, e.g., typical heart rates. We present a sensor-free, deep learning-based navigation that diminishes the need for manual feature engineering or the necessity of prior knowledge compared to previous works. A classifier is trained to estimate the R-wave timepoints in the scan directly from the imaging data. Our approach is evaluated on 3-D protocols for continuous cardiac MRI, acquired in-vivo and free-breathing with single or multiple imaging contrasts. We achieve an accuracy of > 98% on previously unseen subjects, and a well comparable image quality with the state-of-the-art ECG-based reconstruction. Our method enables an ECG-free workflow for continuous cardiac scans with simultaneous anatomic and functional imaging with multiple contrasts. It can be potentially integrated without adapting the sampling scheme to other continuous sequences by using the imaging data for navigation and reconstruction.

In tomographic imaging, anatomical structures are reconstructed by applying a pseudo-inverse forward model to acquired signals. Geometric information within this process is usually depending on the system setting only, i. e., the scanner position or readout direction. Patient motion therefore corrupts the geometry alignment in the reconstruction process resulting in motion artifacts. We propose an appearance learning approach recognizing the structures of rigid motion independently from the scanned object. To this end, we train a siamese triplet network to predict the reprojection error (RPE) for the complete acquisition as well as an approximate distribution of the RPE along the single views from the reconstructed volume in a multi-task learning approach. The RPE measures the motioninduced geometric deviations independent of the object based on virtual marker positions, which are available during training. We train our network using 27 patients and deploy a 21-4-2 split for training, validation and testing. In average, we achieve a residual mean RPE of 0.013mm with an inter-patient standard deviation of 0.022 mm. This is twice the accuracy compared to previously published results. In a motion estimation benchmark the proposed approach achieves superior results in comparison with two state-of-the-art measures in nine out of twelve experiments. The clinical applicability of the proposed method is demonstrated on a motion-affected clinical dataset.

Continuous protocols for cardiac magnetic resonance imaging enable sampling of the cardiac anatomy simultaneously resolved into cardiac phases. To avoid respiration artifacts, associated motion during the scan has to be compensated for during reconstruction. In this paper, we propose a sampling adaption to acquire 2-D respiration information during a continuous scan. Further, we develop a pipeline to extract the different respiration states from the acquired signals, which are used to reconstruct data from one respiration phase. Our results show the benefit of the proposed workflow on the image quality compared to no respiration compensation, as well as a previous 1-D respiration navigation approach.

Recently, Magnetic Resonance Fingerprinting (MRF) was proposed as a quantitative imaging technique for the simultaneous acquisition of tissue parameters such as relaxation times T1 and T2. Although the acquisition is highly accelerated, the state-of-the-art reconstruction suffers from long computation times: Template matching methods are used to find the most similar signal to the measured one by comparing it to pre-simulated signals of possible parameter combinations in a discretized dictionary. Deep learning approaches can overcome this limitation, by providing the direct mapping from the measured signal to the underlying parameters by one forward pass through a network.

Synopsis To perform static cardiac imaging, manual inspection of CINE images is currently necessary to detect a quiescent window within the cardiac cycle. We propose an automated system using two chained Deep Neural Networks to determine localized end-systolic and end-diastolic resting phases. The first network finds a region of interest (e.g. RCA, right or left atrium) and the second determines a quantitative motion curve for this region. Training and evaluation was performed on data from volunteers and patients acquired on different scanners and field strengths and a comparison to manually annotated resting phases showed accuracy in the range of 35ms.

Enforcing geometric consistency of an acquired cone-beam computed tomography scan has been shown to be a promising approach for online geometry calibration and the compensation of rigid patient motion. The approach estimates the motion parameters by solving an optimization problem, where the cost function is the accumulated consistency based on Grangeat’s theorem. In all previous work, this is performed with zero-order optimization methods like the Nelder-Mead algorithm or grid search. We present a derivation of motion gradients enabling the usage of more efficient first-order optimization algorithms for the estimation of rigid patient motion or geometry misalignment. We first present a general formulation of the gradients, and explicitly compute the gradient for the longitudinal patient axis. To verify our results, we compare the presented analytic gradient with a finite difference. In a second experiment we compare the computational demand of the presented gradient with the finite differences. The analytic gradient clearly outperforms the finite differences with a speed up of ∼ 35 %.

Magnetic Resonance Fingerprinting (MRF) is an imaging technique acquiring unique time signals for different tissues. Although the acquisition is highly accelerated, the reconstruction time remains a problem, as the state-of-the-art template matching compares every signal with a set of possible signals. To overcome this limitation, deep learning based approaches, e.g. Convolutional Neural Networks (CNNs) have been proposed. In this work, we investigate the applicability of Recurrent Neural Networks (RNNs) for this reconstruction problem, as the signals are correlated in time. Compared to previous methods based on CNNs, RNN models yield significantly improved results using in-vivo data.

Open gantry C-arm systems that are placed within the interventional room enable 3-D imaging and guidance for stroke therapy without patient transfer. This can profit in drastically reduced time-totherapy, however, due to the interventional setting, the data acquisition is comparatively slow. Thus, involuntary patient motion needs to be estimated and compensated to achieve high image quality. Patient motion results in a misalignment of the geometry and the acquired image data. Consistency measures can be used to restore the correct mapping to compensate the motion. They describe constraints on an idealized imaging process which makes them also sensitive to beam hardening, scatter, truncation or overexposure. We propose a probabilistic approach based on the Student’s t-distribution to model image artifacts that affect the consistency measure without sourcing from motion.

Purpose For a perfectly plane symmetric object, we can find two views—mirrored at the plane of symmetry—that will yield the exact same image of that object. In consequence, having one image of a plane symmetric object and a calibrated camera, we automatically have a second, virtual image of that object if the 3-D location of the symmetry plane is known. Methods We propose a method for estimating the symmetry plane from a set of projection images as the solution of a consistency maximization based on epipolar consistency. With the known symmetry plane, we can exploit symmetry to estimate in-plane motion by introducing the X-trajectory that can be acquired with a conventional short-scan trajectory by simply tilting the acquisition plane relative to the plane of symmetry. Results We inspect the symmetry plane estimation on a real scan of an anthropomorphic human head phantom and show the robustness using a synthetic dataset. Further, we demonstrate the advantage of the proposed method for estimating in-plane motion using the acquired projection data. Conclusion Symmetry breakers in the human body are widely used for the detection of tumors or strokes. We provide a fast estimation of the symmetry plane, robust to outliers, by computing it directly from a set of projections. Further, by coupling the symmetry prior with epipolar consistency, we overcome inherent limitations in the estimation of in-plane motion.

Recently, Magnetic Resonance Fingerprinting (MRF) was proposed as a quantitative imaging technique for the simultaneous acquisition of tissue parameters such as relaxation times T_1 and T_2 . Although the acquisition is highly accelerated, the state-of-the-art reconstruction suffers from long computation times: Template matching methods are used to find the most similar signal to the measured one by comparing it to pre-simulated signals of possible parameter combinations in a discretized dictionary. Deep learning approaches can overcome this limitation, by providing the direct mapping from the measured signal to the underlying parameters by one forward pass through a network. In this work, we propose a Recurrent Neural Network (RNN) architecture in combination with a novel quantile layer. RNNs are well suited for the processing of time-dependent signals and the quantile layer helps to overcome the noisy outliers by considering the spatial neighbors of the signal. We evaluate our approach using in-vivo data from multiple brain slices and several volunteers, running various experiments. We show that the RNN approach with small patches of complex-valued input signals in combination with a quantile layer outperforms other architectures, e.g. previously proposed Convolutional Neural Networks for the MRF reconstruction reducing the error in T_1 and T_2 by more than 80

This work demonstrates the successful application of Deep Learning with phantom and human measurements for the reconstruction in Magnetic Resonance Fingerprinting (MRF). State-of-the-art MRF reconstruction yields quantitative maps of e.g. T and T by acquiring multiple undersampled images with various acquisition parameters, commonly referred to as fingerprints. Every measured fingerprint (per voxel) is compared with a dictionary of simulated fingerprints for possible parameter combinations. This time-consuming step can be replaced with a neural network, which directly predicts the parameters from a fingerprint. This was previously shown with simulated data. Here, we extend this approach to real measurements.

Purpose To develop a bipolar multi‐echo MRI method for the accurate estimation of the adipose tissue fatty acid composition (FAC) using clinically relevant protocols at clinical field strength. Methods The proposed technique jointly estimates confounding factors (field map, , eddy‐current phase) and triglyceride saturation state parameters by fitting multi‐echo gradient echo acquisitions to a complex signal model. The noise propagation behavior was improved by applying a low‐rank enforcing denoising technique and by addressing eddy‐current‐induced phase discrepancies analytically. The impact of the total echo train duration on the FAC parameter map accuracy was analyzed in an oil phantom at 3T. Accuracy and reproducibility assessment was based on in vitro oil phantom measurements at two field strengths (3T and 1.5T) and with two different protocols. Repeatability was assessed in vivo in patients ( n = 8) with suspected fatty liver disease using test–retest acquisitions in the abdominal subcutaneous, perirenal and mesenteric fat depots. Results Echo train readout durations of at least five times the conventional in‐phase time were required for accurate FAC estimation in areas of high fat content. In vitro, linear regression and Bland–Altman analyses demonstrated strong ( r > 0.94) and significant ( P ≪ 0.01) correlations between measured and reference FACs for all acquisitions, with smaller overall intercepts and biases at 3T compared to 1.5T. In vivo, reported mean absolute differences between test and retest were 1.54%, 3.31%, and 2.63% for the saturated, mono‐unsaturated, and poly‐unsaturated fat component, respectively. Conclusions Accurate and reproducible MRI‐based FAC quantification within a breath‐hold is possible at clinical field strengths.

Synopsis We present a free-breathing multi-contrast 3-D cardiac CINE acquisition and reconstruction technique based on Compressed Sensing. Inversion pulses were repeatedly applied during a continuous acquisition to sample contrastand cardiac-resolved 3-D data, while a self-navigation method was applied for respiratory gating. Validation was performed in a phantom, showing recovery curves and T * maps in good correlation to known T values for the phantom as well as a MOLLI reference measurement. Feasibility for in-vivo application was demonstrated in a healthy volunteer.

Pattern Recognition Lab, Department of Computer Science, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany, MR Application Predevelopment, Siemens Healthcare GmbH, Erlangen, Germany, MR R&D Collaborations, Siemens Healthineers, Cary, NC, United States, Radiology, Duke University Medical Center, Durham, NC, United States, Center for Advanced Magnetic Resonance Development, Duke University Medical Center, Durham, NC, United States

Pattern Recognition Lab, Department of Computer Science, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany, MR Application Predevelopment, Siemens Healthcare GmbH, Erlangen, Germany, MR R&D Collaborations, Siemens Healthineers, Cary, NC, United States, Radiology, Duke University Medical Center, Durham, NC, United States, Center for Advanced Magnetic Resonance Development, Duke University Medical Center, Durham, NC, United States

The purpose of this work is to evaluate methods from deep learning for application to Magnetic Resonance Fingerprinting (MRF). MRF is a recently proposed measurement technique for generating quantitative parameter maps. In MRF a non-steady state signal is generated by a pseudo-random excitation pattern. A comparison of the measured signal in each voxel with the physical model yields quantitative parameter maps. Currently, the comparison is done by matching a dictionary of simulated signals to the acquired signals. To accelerate the computation of quantitative maps we train a Convolutional Neural Network (CNN) on simulated dictionary data. As a proof of principle we show that the neural network implicitly encodes the dictionary and can replace the matching process.