ECG gated breath-hold cine MRI with b-SSFP (1,2) readout is considered the gold standard test for assessment of cardiac function. However, it has two noted limitations: (i) breath-holding requirement and (ii) sensitivity to arrhythmia. Real-time cine MRI is a solution to these challenges, but commercially available methods are still limited by poor data acquisition efficiency. To date, a variety of acceleration methods have been proposed, including dynamic parallel imaging (3,4), k-t parallel imaging methods such as BLAST/SENSE (5), and compressed sensing (CS) (6-9), with each method having advantages and disadvantages.

In this study, we explore CS as a vehicle to accelerate real-time cine MRI and compare its performance between Cartesian and radial k-space sampling schemes. Cartesian acquisition has several advantages such as a reduced sensitivity to gradient imperfections and capability to use an asymmetric field of view (FOV) adopted to the human chest. Radial has its own advantages, including better motion properties (10), capability to retrospectively determine temporal resolution when using non-repeated sampling schemes such as golden angle ratio (11), and higher signal to noise ratio (SNR) due to higher sampling of the center of k-space. To date, no study has systematically compared real-time cine MRI between Cartesian and radial k-space sampling schemes. We sought to compare their performance in patients and volunteers.

(Theoretical Evaluation) Radial and Cartesian sampling patterns were compared using analysis of point spread functions (PSF). For a fair comparison, both radial and Cartesian sampling patterns were made with 144 frequency-encoding points and 12 phase-encoding lines per cardiac phase, and over 30 cardiac frames, as shown in Figure 1. Fourier transforms were performed to obtain PSFs in each 2D space (x-y, y-f, and x-f) at midpoint of the third dimension, for ease in evaluation. PSFs were analyzed to calculate the relative energy of the central spire to the side lobes (energy ratio = central spire / side lobes, where higher number translates to more incoherent aliasing artifacts).

(Prospective Evaluation) We compared the performance of three free breathing real-time cine MRI pulse sequences with b-SSFP readouts in a group of 15 patients (9 male, average age 65±14, six female, average age 58±10) diagnosed with arrhythmias and 1 healthy volunteer: (i) threefold accelerated using TGRAPPA (4); (ii) twelvefold-accelerated using CS (12,13) with Cartesian sampling, and (iii) twelvefold-accelerated using CS with radial sampling with tiny golden angles (14). Patients and volunteer were scanned on a combination of 3T (Verio, Trio) and 1.5T (Espree, Avanto) whole-body MRI scanners (Siemens). Table 1 summarizes relevant imaging parameters for the 3 pulse sequences. Note that the radial acquisition had slightly higher spatial resolution because it enables smaller field of view (i.e., oversampling in the frequency-encoding direction) while avoiding considerable fold-over aliasing artifacts.

(Image Analysis) The diagnostic quality of the three different cine MRI methods was graded by two non-invasive cardiologists. A mid-ventricular short-axis slice was chosen from each pulse sequence for visual assessment of diagnostic quality. Forty-two (14 from each pulse sequence) cine datasets were pooled and randomized for blinded analysis. The readers independently scored the following four categories using a Likert scale: image quality (e.g., endocardial-edge definition), temporal fidelity of wall motion, artifact level, and apparent noise level. Averaging over two readers, a score of ≥ 3.0 was considered to be diagnostically acceptable.

Energy ratio calculations are summarized in Figure 1. The radial sampling pattern had a larger energy ratio in every 2D space except for x-f, where the Cartesian sampling pattern was fully sampled. When the energy ratio values of all three 2D spaces were averaged, the radial pattern had a higher mean energy ratio (1.65) when compared to the mean Cartesian energy ratio (1.03).

Figure 2 shows representative images of two different patients diagnosed with atrial fibrillation on 1.5 and 3T scanners, where radial images consistently produced better image quality than Cartesian CS and TGRAPPA images. The mean scores are summarized in Table 2. According to the Kruskal-Wallis test, all 4 categories were significantly different (p < 0.05). With TGRAPPA as the control, Cartesian CS was significantly different only for the temporal fidelity score, whereas radial CS was significantly different for all 4 categories. Only radial CS resulted in acceptable scores for all 4 categories.

This study demonstrates that accelerated real-time cine MRI with radial k-space sampling produces superior image quality than accelerated real-time cine MRI with Cartesian k-space sampling. Future studies could evaluate the clinical utility of accelerated real-time cine MRI with radial k-space sampling for assessment of cardiac function in pediatric patients or patients with arrhythmia.