Automated Protocoling for MRI Exams-Challenges and Solutions.

Automated protocoling for MRI examinations is an amendable target for workflow automation with artificial intelligence. However, there are still challenges to overcome for a successful and robust approach. These challenges are outlined and analyzed in this work. Through a literature review, we analyzed limitations of currently published approaches for automated protocoling. Then, we assessed these limitations quantitatively based on data from a private radiology practice. For this, we assessed the information content provided by the clinical indication by computing the overlap coefficients for the sets of ICD-10-coded admitting diagnoses of different MRI protocols. Additionally, we assessed the heterogeneity of protocol trees from three different MRI scanners based on the overlap coefficient, on MRI protocol and sequence level. Additionally, we applied sequence name standardization to demonstrate its effect on the heterogeneity assessment, i.e., the overlap coefficient, of different protocol trees. The overlap coefficient for the set of ICD-10-coded admitting diagnoses for different protocols ranges from 0.14 to 0.56 for brain/head MRI exams and 0.04 to 0.57 for spine exams. The overlap coefficient across the set of sequences used at two different scanners increases when applying sequence name standardization (from 0.81/0.86 to 0.93). Automated protocoling for MRI examinations has the potential to reduce the workload for radiologists. However, an automated protocoling approach cannot be solely based on admitting diagnosis as it does not provide sufficient information. Moreover, sequence name standardization increases the overlap coefficient across the set of sequences used at different scanners and therefore facilitates transfer learning.

Enhanced Magnetic Resonance Image Synthesis with Contrast-Aware Generative Adversarial Networks.

A magnetic resonance imaging (MRI) exam typically consists of the acquisition of multiple MR pulse sequences, which are required for a reliable diagnosis. With the rise of generative deep learning models, approaches for the synthesis of MR images are developed to either synthesize additional MR contrasts, generate synthetic data, or augment existing data for AI training. While current generative approaches allow only the synthesis of specific sets of MR contrasts, we developed a method to generate synthetic MR images with adjustable image contrast. Therefore, we trained a generative adversarial network (GAN) with a separate auxiliary classifier (AC) network to generate synthetic MR knee images conditioned on various acquisition parameters (repetition time, echo time, and image orientation). The AC determined the repetition time with a mean absolute error (MAE) of 239.6 ms, the echo time with an MAE of 1.6 ms, and the image orientation with an accuracy of 100%. Therefore, it can properly condition the generator network during training. Moreover, in a visual Turing test, two experts mislabeled 40.5% of real and synthetic MR images, demonstrating that the image quality of the generated synthetic and real MR images is comparable. This work can support radiologists and technologists during the parameterization of MR sequences by previewing the yielded MR contrast, can serve as a valuable tool for radiology training, and can be used for customized data generation to support AI training.

MR-contrast-aware image-to-image translations with generative adversarial networks.

PURPOSE: A magnetic resonance imaging (MRI) exam typically consists of several sequences that yield different image contrasts. Each sequence is parameterized through multiple acquisition parameters that influence image contrast, signal-to-noise ratio, acquisition time, and/or resolution. Depending on the clinical indication, different contrasts are required by the radiologist to make a diagnosis. As MR sequence acquisition is time consuming and acquired images may be corrupted due to motion, a method to synthesize MR images with adjustable contrast properties is required. METHODS: Therefore, we trained an image-to-image generative adversarial network conditioned on the MR acquisition parameters repetition time and echo time. Our approach is motivated by style transfer networks, whereas the "style" for an image is explicitly given in our case, as it is determined by the MR acquisition parameters our network is conditioned on. RESULTS: This enables us to synthesize MR images with adjustable image contrast. We evaluated our approach on the fastMRI dataset, a large set of publicly available MR knee images, and show that our method outperforms a benchmark pix2pix approach in the translation of non-fat-saturated MR images to fat-saturated images. Our approach yields a peak signal-to-noise ratio and structural similarity of 24.48 and 0.66, surpassing the pix2pix benchmark model significantly. CONCLUSION: Our model is the first that enables fine-tuned contrast synthesis, which can be used to synthesize missing MR-contrasts or as a data augmentation technique for AI training in MRI. It can also be used as basis for other image-to-image translation tasks within medical imaging, e.g., to enhance intermodality translation (MRI → CT) or 7 T image synthesis from 3 T MR images.

Robust cardiac BOLD MRI using an fMRI-like approach with repeated stress paradigms.

PURPOSE: Use repeated stress paradigms and an approach taken from neurological blood oxygen level dependent (BOLD) functional MRI (fMRI) to derive robust cardiac BOLD measurements. METHODS: Multiple-repetition, single-shot, electrocardiograph-triggered, T2-prepared BOLD balanced steady-state free precession was performed during repeated long breath-holds in 13 volunteers. Nonrigid motion correction was applied to the continuously acquired data and it was analyzed with a general linear model (GLM) taking into account the effects of the breath-hold duration, RR interval, motion, and baseline variations. Both voxel- and region of interest-based analyses were performed. RESULTS: The GLM model was able to isolate the component of the BOLD signal arising from the breath-holds and separate it from the background effects due to the changing heart rate and motion. A significant (P<0.05) BOLD signal increase was observed in the myocardium of healthy volunteers. CONCLUSION: Using a recent elastic motion correction algorithm and adapted acquisition techniques, it was possible to apply fMRI-like strategies for cardiac BOLD MRI in volunteers and derive robust BOLD measurements. The observed slight but significant oxygenation increase in the myocardium of volunteers might be explained by the vasodilator effect of increased CO2 concentration under apnea. Detection of such small physiological changes in volunteers performing breath-holds demonstrates that the method could have potential in identifying low oxygenation regions in the myocardium of patients during stress tests.

Magnetic resonance 4D flow reveals unusual hemodynamics associated with aneurysm formation and a possible cause of cryptogenic stroke in a patient with aortic dissection.

Four-dimensional flow is a magnetic resonance technology that has undergone significant technical improvements in recent years. With increasingly rapid acquisition times and new postprocessing tools, it can provide a tool for demonstrating and visualizing cardiovascular flow phenomena, which may offer new insights into disease. We present an interesting clinical case in which 4-dimensional flow demonstrates potential etiologies for 2 interesting phenomena in the same patient: (1) development of an unusual aneurysm and (2) cryptogenic stroke.

Phase-sensitive inversion recovery for myocardial T1 mapping with motion correction and parametric fitting.

The assessment of myocardial fibrosis and extracellular volume requires accurate estimation of myocardial T1 s. While image acquisition using the modified Look-Locker inversion recovery technique is clinically feasible for myocardial T1 mapping, respiratory motion can limit its applicability. Moreover, the conventional T1 fitting approach using the magnitude inversion recovery images can lead to less stable T1 estimates and increased computational cost. In this article, we propose a novel T1 mapping scheme that is based on phase-sensitive image reconstruction and the restoration of polarity of the MR signal after inversion. The motion correction is achieved by registering the reconstructed images after background phase removal. The restored signal polarity of the inversion recovery signal helps the T1 fitting resulting in improved quality of the T1 map and reducing the computational cost. Quantitative validation on a data cohort of 45 patients proves the robustness of the proposed method against varying image contrast. Compared to the magnitude T1 fitting, the proposed phase-sensitive method leads to less fluctuation in T1 estimates.

Myocardial T2 mapping with respiratory navigator and automatic nonrigid motion correction.

Quantitative T2 mapping was recently shown to be superior to T2-weighted imaging in detecting T2 changes across myocardium. Pixel-wise T2 mapping is sensitive to misregistration between the images used to generate the parameter map. In this study, utility of two motion-compensation strategies-(i) navigator gating with prospective slice correction and (ii) nonrigid registration-was investigated for myocardial T2 mapping in short axis and horizontal long axis views. Navigator gating provides respiratory motion compensation, whereas registration corrects for residual cardiac and respiratory motion between images; thus, the two strategies provided complementary functions. When these were combined, respiratory-motion-induced T2 variability, as measured by both standard deviation and interquartile range, was comparable to that in breath-hold T2 maps. In normal subjects, this combined motion-compensation strategy increased the percentage of myocardium with T2 measured to be within normal range from 60.1% to 92.2% in short axis and 62.3% to 92.7% in horizontal long axis. The new motion-compensated T2 mapping technique, which combines navigator gating, prospective slice correction, and nonrigid registration to provide through-plane and in-plane motion correction, enables a method for fully automatic and robust free-breathing T2 mapping.

T2-weighted cardiac MR assessment of the myocardial area-at-risk and salvage area in acute reperfused myocardial infarction: comparison of state-of-the-art dark blood and bright blood T2-weighted sequences.

PURPOSE: To compare different state-of-the-art T2-weighted (T2w) imaging sequences combined with late gadolinium enhancement (LGE) for myocardial salvage area (MSA) assessment by cardiac magnetic resonance (CMR). T2w imaging has been used to assess the myocardial area at risk (AAR) in acute myocardial infarction (AMI) patients, but its clinical application is challenging due to technical and physical limitations. MATERIALS AND METHODS: Thirty patients with reperfused AMI underwent complete CMR imaging 2-5 days after hospital admission. Myocardial AAR and MSA were quantified on four different T2w sequences: (a) free-breathing T2-prepared single-shot balanced steady-state free precession (T2p_ssbSSFP); (b) breathhold T2-weighted acquisition for cardiac unified T2 edema (ACUTE); (c) breathhold T2w dark-blood inversion recovery turbo-spin echo (IR-TSE) (short-term inversion recovery: STIR); and (d) free-breathing high-resolution T2 dark-blood navigated BLADE. The diagnostic performance of each technique was also assessed. RESULTS: Quantitative analysis showed significant differences in myocardial AAR extent as quantified by the four T2w sequences (P < 0.05). There were also significant differences in sensitivity, specificity and overall diagnostic performance. CONCLUSION: Detection and quantification of AAR, and thus of MSA, by T2wCMR in reperfused AMI patients varied significantly between different T2w sequences in the same clinical setting.

Non-rigid registration and KLT filter to improve SNR and CNR in GRE-EPI myocardial perfusion imaging.

The purpose of the study was to evaluate the effect of motion compensation by non-rigid registration combined with the Karhunen-Loeve Transform (KLT) filter on the signal to noise (SNR) and contrast-to-noise ratio (CNR) of hybrid gradient-echo echoplanar (GRE-EPI) first-pass myocardial perfusion imaging. Twenty one consecutive first-pass adenosine stress perfusion MR data sets interpreted positive for ischemia or infarction were processed by non-rigid Registration followed by KLT filtering. SNR and CNR were measured in abnormal and normal myocardium in unfiltered and KLT filtered images following non-rigid registration to compensate for respiratory and other motions. Image artifacts introduced by filtering in registered and nonregistered images were evaluated by two observers. There was a statistically significant increase in both SNR and CNR between normal and abnormal myocardium with KLT filtering (mean SNR increased by 62.18% ± 21.05% and mean CNR increased by 58.84% ± 18.06%; p = 0.01). Motion correction prior to KLT filtering reduced significantly the occurrence of filter induced artifacts (KLT only-artifacts in 42 out of 55 image series vs. registered plus KLT-artifacts in 3 out of 55 image series). In conclusion the combination of non- rigid registration and KLT filtering was shown to increase the SNR and CNR of GRE-EPI perfusion images. Subjective evaluation of image artifacts revealed that prior motion compensation significantly reduced the artifacts introduced by the KLT filtering process.

Motion correction for myocardial T1 mapping using image registration with synthetic image estimation.

Quantification of myocardial T1 relaxation has potential value in the diagnosis of both ischemic and nonischemic cardiomyopathies. Image acquisition using the modified Look-Locker inversion recovery technique is clinically feasible for T1 mapping. However, respiratory motion limits its applicability and degrades the accuracy of T1 estimation. The robust registration of acquired inversion recovery images is particularly challenging due to the large changes in image contrast, especially for those images acquired near the signal null point of the inversion recovery and other inversion times for which there is little tissue contrast. In this article, we propose a novel motion correction algorithm. This approach is based on estimating synthetic images presenting contrast changes similar to the acquired images. The estimation of synthetic images is formulated as a variational energy minimization problem. Validation on a consecutive patient data cohort shows that this strategy can perform robust nonrigid registration to align inversion recovery images experiencing significant motion and lead to suppression of motion induced artifacts in the T1 map.

Motion compensated magnetic resonance reconstruction using inverse-consistent deformable registration: application to real-time cine imaging.

Patient motion is a major limitation for magnetic resonance imaging. Recent theoretical advances incorporate explicit rigid and non-rigid motion compensation into conventional image reconstruction for multi-shot acquisitions and recover motion-free images by solving a general matrix inversion problem. Although the theory has been established, applications are rare due to the challenges of estimating motion field for every pixel of every shot. In this paper we propose a method to overcome this difficulty using the inverse-consistent deformable registration supplying both forward and backward deformations for matrix inversion. We further extend this framework for multi-coil motion compensated image reconstruction using the eigen-mode analysis. Both simulations and in vivo studies demonstrate the effectiveness of our approach.

Automatic view planning for cardiac MRI acquisition.

Conventional cardiac MRI acquisition involves a multi-step approach, requiring a few double-oblique localizers in order to locate the heart and prescribe long- and short-axis views of the heart. This approach is operator-dependent and time-consuming. We propose a new approach to automating and accelerating the acquisition process to improve the clinical workflow. We capture a highly accelerated static 3D full-chest volume through parallel imaging within one breath-hold. The left ventricle is localized and segmented, including left ventricle outflow tract. A number of cardiac landmarks are then detected to anchor the cardiac chambers and calculate standard 2-, 3-, and 4-chamber long-axis views along with a short-axis stack. Learning-based algorithms are applied to anatomy segmentation and anchor detection. The proposed algorithm is evaluated on 173 localizer acquisitions. The entire view planning is fully automatic and takes less than 10 seconds in our experiments.

Cardiac anchoring in MRI through context modeling.

Cardiac magnetic resonance imaging (MRI) has advanced to become a powerful diagnostic tool in clinical practice. Robust and fast cardiac modeling is important for structural and functional analysis of the heart. Cardiac anchors provide strong cues to extract morphological and functional features for diagnosis and disease monitoring. We present a fully automatic method and system that is able to detect these cues. The proposed approach explores expert knowledge embedded in a large annotated database. Exemplar cues in our experiments include left ventricle (LV) base plane and LV apex from long-axis images, and right ventricle (RV) insertion points from short-axis images. We evaluate the proposed approach on 8304 long-axis images from 188 patients and 891 short-axis images from 338 patients that are acquired from different vendors. In addition, another evaluation is conducted on an independent 7140 images from 87 patient studies. Experimental results show promise of the proposed approach.

Unsupervised inline analysis of cardiac perfusion MRI.

In this paper we first discuss the technical challenges preventing an automated analysis of cardiac perfusion MR images and subsequently present a fully unsupervised workflow to address the problems. The proposed solution consists of key-frame detection, consecutive motion compensation, surface coil inhomogeneity correction using proton density images and robust generation of pixel-wise perfusion parameter maps. The entire processing chain has been implemented on clinical MR systems to achieve unsupervised inline analysis of perfusion MRI. Validation results are reported for 260 perfusion time series, demonstrating feasibility of the approach.

Combining registration and minimum surfaces for the segmentation of the left ventricle in cardiac cine MR images.

This paper describes a system to automatically segment the left ventricle in all slices and all phases of cardiac cine magnetic resonance datasets. After localizing the left ventricle blood pool using motion, thresholding and clustering, slices are segmented sequentially. For each slice, deformable registration is used to align all the phases, candidates contours are recovered in the average image using shortest paths, and a minimal surface is built to generate the final contours. The advantage of our method is that the resulting contours follow the edges in each phase and are consistent over time. We demonstrate using 19 patient examples that the results are very good. The RMS distance between ground truth and our segmentation is only 1.6 pixels (2.7 mm) and the Dice coefficient is 0.89.

Evaluation of rigid and non-rigid motion compensation of cardiac perfusion MRI.

Although the evaluation of cardiac perfusion using MRI could be of crucial importance for the diagnosis of ischemic heart diseases, it is still not a routinely used technique. The major difficulty is that MR perfusion images are often corrupted by inconsistent myocardial motion. Although motion compensation methods have been studied throughout the past decade, no clinically accepted solution has emerged. This is partly due to the lack of comprehensive validation. To address this deficit we collected a large multi-centre MR perfusion dataset and used this to characterize typical myocardial motion and confirmed that under clinically relevant conditions motion correction is a frequent requirement (67% of all 586 cases). We then developed a proposed solution which includes both rigid/affine and the non-rigid image registration. Quantitative validation has been conducted using 6 different statistics to provide a comprehensive evaluation, showing the proposed techniques to be highly robust to different myocardial anatomy and motion patterns as well as to MR imaging acquisition parameters.

Acquisition-related morphological variability in structural MRI.

RATIONALE AND OBJECTIVES: Significant effort has been spent during the past decades to develop innovative image-processing algorithms and improve existing methods in terms of precision, reproducibility, and computational efficiency, but relatively little research was undertaken to find out the extent to which the validity of results obtained with these methods is limited by inherent imperfections of the input images. This observation is especially true for magnetic resonance imaging (MRI)-based morphometry, which aims at precise and highly reproducible determination of geometric properties of anatomic structures, although MRI images are geometrically distorted. MATERIALS AND METHODS: A method for characterization of site-specific geometric distortions and results of a long-term study designed to find the extent to which imperfections in the data-acquisition process limit the reliable detection of subtle morphological changes in MRI data acquired with state-of-the-art scanners are presented. Because of the long-term character of the study, results include effects resulting from limited hardware stability, as well as from imperfections in patient repositioning. RESULTS: Maximal relative morphological changes detected in our phantom data series were 1.0 mm positional and 2.0% volumetric difference (relative to a 7600-mm3 cuboid) in a subvolume relevant for whole-brain morphometry. Morphological variability was even greater for human volunteer data (up to 5% in local gray matter volume) because of movements during scan, natural morphological variability, and a presumably less precise segmentation procedure. CONCLUSION: Imperfections in the MRI data-acquisition process in combination with practical limitations in patient repositioning can substantially confound studies of subtle morphological changes.

Acquisition-related limitations in MRI based morphometry.

Although significant effort has been spent over the past decades to develop innovative image processing algorithms and to improve existing methods in terms of precision, reproducibility and computational efficiency, relatively few research was undertaken to find out to what extent the validity of results obtained with these methods is limited by inherent imperfections of the input images. This observation is especially true for MRI based morphometry, which aims at precise and highly reproducible determination of geometrical properties of anatomical structures despite the fact that MR images are geometrically distorted. We here present (a) a method for characterization of site-specific geometrical distortions and (b) the results of a long term study designed to find out how precisely geometrical properties and morphological changes of brain structures can, in principle, be detected in images acquired with MRI scanners. Due to the long-term character of our study, our findings include effects resulting from limited hardware stability as well as from variations in patient positioning. Our results show that these effects can be strong enough to substantially confound MRI studies of small morphological changes.