In Cartesian Cardiac MRI, respiratory motion leads to distinct ghosting artifacts. This necessitates respiratory navigators or other means of motion selection/correction also for Cartesian Tissue Phase Mapping (TPM) [1]. With radial MRI, however, motion artifacts result in image blurring rather than ghosting, which might allow using all respiratory states for reconstruction. The aim of this study is to investigate the influence of respiratory motion on velocities obtained from radial Tissue Phase Mapping MRI.

Acquisition: In 10 healthy volunteers, a retrospectively triggered radial golden angle black-blood TPM sequence with 4-point balanced (Hadamard) velocity encoding and VENC = 30 cm/s was applied in 3 short axis slices. Acquisition parameters were: TR/TE = 5.6 ms / 3.4 ms, flip angle α = 15°, resolution = 1.6 x 1.6 mm², slice thickness = 8 mm, and FOV = 340 x 340 mm².

Reconstruction: Raw data was exported and processed with Matlab. An image-based self-gating signal [2] was generated for respiratory gating. The respiration signal was divided into 6 respiration bins with accepting window widths adapted to ensure similar undersampling in each bin. The bin with the most occurring respiratory position was defined as the reference bin. Bins with more than 6 mm acceptance window were rejected to avoid risk of intra-bin motion. The following cine reconstructions were performed:

· FREE: no respiratory gating (accept all data);

· REF: only use reference bin;

· AVG: reconstruct respiration bins separately, then average;

· MC: as AVG, but perform motion correction (affine image registration [3] in ROI around heart) before averaging.

Reconstructions were done by gridding (FREE; low undersampling) or iteratively (all others) by GRASP [4].

Analysis: Image quality and velocity measures for the different reconstructions were compared by Wilcoxon signed-rank tests with Bonferroni correction for multiple testing and p-values below 5% were considered significant.

Figure 1 shows magnitude and velocity images for the different reconstruction methods. Visually, REF has less SNR than the other reconstructions. FREE and AVG appear slightly blurred, whereas MC is as sharp as REF. Velocity analysis can be appreciated in figure 2. Segmental velocities over the cardiac cycle are similar for all reconstruction methods, with only slightly spatial deviations in REF. Global velocities are also highly similar over the cardiac cycle. Slight reduction of peaks velocities, however, are apparent compared to REF, especially in FREE. Velocity peak times between the different reconstruction methods are neither different in the segmental nor in the global analysis.

Statistical analysis (Figure 3) confirms the visual impressions. SNR is improved in all methods compared to REF due to the usage of all acquired rather than only selected data. Sharpness and image contrast are decreased in FREE and AVG, but similar in MC to REF. Only some of the velocity peaks are reduced in FREE compared to REF. Overall agreement of velocities with REF as measured via velocity correlation and RMSE is higher in AVG and MC than in FREE.

While consideration of respiratory motion seems to be necessary in Cartesian TPM, free-breathing reconstruction is feasible for radial TPM without significant influence on the resulting velocities. The small bias towards reduced velocity peaks might be decreased via respiratory binning and (motion-compensated) averaging, thus keeping the 100% navigator efficiency of FREE compared to typically lower acceptance rates of 40%-60% in gated acquisitions (REF).