Our study protocol was approved by the local Institutional Review Board and written informed consent was obtained from all patients. The study was conducted in compliance with the Health Insurance Portability and Accountability Act guidelines. Patients (n = 25) with known thoracic aorta dilation were prospectively enrolled for a research study between July 2017 and September 2018. Further inclusion criteria were: (1) > 18 years of age; (2) previous clinically indicated aorta, chest, triple-rule-out, or pulmonary embolism computed tomography examination; and (3) willing to comply with all study procedures and provide written informed consent. General magnetic resonance exclusion criteria were applied to patient selection. Patient’s demographics were obtained by medical record chart review.
Image acquisition was performed on a 1.5-T system (Magnetom Avanto DOT, Siemens Healthcare, Erlangen, Germany). Patients were scanned head-first in a supine position. A multi-channel spine phased-array radiofrequency coil with 24 elements integrated into the patient table and a 6 element, 6-channel phased-array body coil were used for signal reception. Acquisitions were electrocardiographically gated. The entire protocol was performed in a free-breathing fashion. Following the initial scout images, a free-breathing two-dimensional balanced SSFP cine image set in a parasagittal long-axis view of the left ventricle was acquired using the following typical parameters: repetition time/echo time, 2.3/1.1 ms; field of view 220–340 mm2; matrix 1922; number of segments 15; reconstructed phases 25; temporal resolution 45 ms; flip angle 77°; number of excitations 3; and parallel acquisition acceleration factor 2. Cine image data were used to determine the optimal mid-diastolic timing for the whole-heart MRA.
Whole-heart MRA was performed using a prototype fat-saturated and T2-prepared pulse sequence that employs the 3D radial trajectory [7, 22]. A coronal saturation slab was placed at the level of the anterior chest wall and the following imaging parameters were used: repetition time/echo time 3.1/1.5 ms; field of view 320 × 320 × 320 mm3; reconstructed voxel size, 1.7 × 1.7 × 1.7 mm3; matrix 1923; flip angle 115°; and bandwidth, 898 Hz/pixel. An acceleration factor of 5 with respect to the Nyquist sampling for 3D radial imaging was applied . The acquisition was performed in free-breathing and the following two approaches were used to address respiratory motion within the reconstruction.
Respiratory motion-corrected approach
For the self-navigation approach, respiratory motion was extracted by cross-correlating the automatically segmented blood pool of the 1D Fourier transform of a readout along the superior-inferior direction acquired consistently at the beginning of each heartbeat. The detected 1D superior-inferior respiratory displacement was then used for correcting each readout before the gridding operation. The correction was performed by applying a phase shift to all k-space radial readouts and was adapted for the polar orientation of each readout according to the spiral phyllotaxis pattern . This reconstruction has been implemented inline at the scanner and takes about 1–2 min. Further details about the algorithm employed for motion correction were described by Piccini et al. .
Respiratory motion-resolved approach
For the respiratory motion-resolved approach, the same raw data of the whole-heart MRA acquisition were exported and processed on a dedicated workstation using an adaptation of the previously described framework  implemented in MATLAB 2015a (MathWorks, Natick, Massachusetts, USA). Using a respiratory signal extracted directly from the image data, individual readouts of the 3D radial acquisition were binned according to their respiratory phase . The resultant series of motion-resolved undersampled images were then reconstructed using an XD-GRASP algorithm , which aims at exploiting the intrinsic similarities between distinct respiratory phases (or motion states) of the whole-heart acquisition to perform a compressed sensing reconstruction along the respiratory motion dimension. To achieve this, first, the acquired readouts are separated and grouped according to the respiratory phase they belong to. Subsequently, a k-t sparse SENSE iterative reconstruction is performed, where the temporal domain is represented by the different respiratory phases. Out of the four reconstructed respiratory phases, the end-expiratory phase was selected in all datasets for the subsequent analyses. The motion resolved reconstruction has not been implemented inline and takes between 15–30 min using high-end computers.
Respiratory motion-corrected and motion-resolved MRA reconstructions were randomized and independently reviewed by two readers (with 1 and 11 years of experience in cardiovascular imaging, respectively) on a dedicated workstation (Aquarius iNtuition Edition v4.4.12, TeraRecon, Inc., Foster City, CA, USA). Standard axial, coronal, and sagittal planes were used to generate multi-planar reformats (MPR), but readers were allowed to use curved MPR or maximum intensity projection (MIP) series according to their preference. The presence of artifacts was noted.
The overall image quality was subjectively rated independently by each reader on a 5-point Likert-scale: (1) vascular anatomy not assessable due to severe image artifacts and/or poor contrast, (2) vascular anatomy assessable despite severe image artifacts and/or poor contrast, (3) acceptable image quality with artifacts and/or limited contrast, (4) good image quality with minor artifacts and/or good contrast, and (5) excellent image quality without artifacts and excellent contrast. Signal homogeneity in the intra-aortic blood pool was rated using a 3-point Likert-scale as (1) inhomogeneity affecting diagnosis, (2) subtle inhomogeneity with no effect on diagnosis, and (3) excellent homogeneity. Image sharpness was evaluated on a 3-point scale as follows: (1) motion affecting diagnosis, (2) motion with no effect on diagnosis, and (3) no significant motion. Finally, diagnostic confidence was also rated by each reader using a 3-point scale as (1) low reader confidence, (2) marginal reader confidence, and (3) high reader confidence.
Readers visualized seven standard anatomical levels of the thoracic aorta using a double oblique technique as follows: sinuses of Valsalva, sinotubular junction, mid ascending aorta, proximal aortic arch, mid aortic arch, proximal descending aorta, and mid descending aorta . At each level, the signal intensity ratio (SIR) between the intravascular signal and the surrounding lung tissue was calculated. Regions of interest were placed in the center of the aorta. Aorta blood pool signal homogeneity was quantified by measuring the standard deviation of blood signal as a function of distance along the thoracic aorta on centerline reconstructions. Finally, the sharpness of the right coronary artery was quantitatively evaluated using a dedicated prototype application (Soap-Bubble, John’s Hopkins University, Baltimore, MD, USA) . This software examines a user-assisted definition of a curved subvolume enclosed in the 3D MRA dataset and measures the magnitude of local change in signal intensity at the vessel borders in the reformatted image. The resulting vessel edge value indicates quantitative sharpness, whereas a value of 100% refers to an abrupt signal intensity change at the vessel border, and lower values are associated with lower vessel sharpness . Measurements were taken in the proximal segment of the right coronary artery, being the most sensitive to motion .
Statistical analysis was performed using SPSS v213 (IBM Corporation, Armonk, NY, USA). Categorical variables are represented as total number and percentages, and continuous variables as mean ± standard deviation or median (interquartile range), depending on their distribution (tested with Shapiro Wilkes). Subjective image quality scores were compared between the respiratory motion-corrected and motion-resolved techniques using the Wilcoxon signed-rank test, and the McNemar test was used to compare the presence of image artifacts. Interclass correlations (ICC) were used to assess the absolute agreement between readers and were interpreted as follows: < 0.2, poor; 0.2–0.4, acceptable; 0.41–0.6, moderate; 0.61–0.8, good; and > 0.8, excellent. Objective image quality measures were compared using a paired Student’s t test. A p value < 0.05 was considered significant.