This work compares two center-out UTE trajectories providing similar contrast. The classic radial UTE [1] that is realized by a straight radial center-out read-out (RO), followed by a rephasing gradient along the same trajectory before spoiling. The second trajectory takes advantage of the rephasing time and replaces the reversed gradients in radial UTE by a rosette petal parametrized by k t = k x t + i k y t = k max ⁡ sin ω t e i t , (1) where ω accounts for the general shape of the rosette and is chosen from the interval ω ∈ R . For t ∈ [ 0 , π / ω ] the rosette reduces to a single petal, starting and ending in k 0 . By doing so, the k-space coverage by read-out can be highly increased with almost negligible increase in repetition time. The trajectory design is described in further detail in [12]. During playing out the entire petal, data is acquired, however, if solely the first half of the petal is used for image reconstruction, radial UTE-like contrast can be achieved. The full petal is used for the reconstruction of preliminary images applied for gating as further described in section 2.2.

A parametrization of the SPR trajectory of ω = 5 / 3 in Equation 1 was chosen for this publication. The trajectory was calculated in MATLAB (MathWorks, Natick, Massachusetts, United States) and respective gradient waveforms derived by minimizing read-out time considering a maximum gradient strength of 20 mT/m, as well as maximum slew rate of 150 T/m/s. Total read-out duration for SPR was t acq, SPR = 1.01  ms, leading to an effective acquisition time of t acq, SPR, eff = 0.51  ms for the center-out part used for final image reconstruction.

Images were acquired on a Philips 3 T whole body MRI (Ingenia 3.0 T CX, Philips Healthcare, Best, Netherlands) using a 16ch DStream torso coil (Philips Healthcare, Best, Netherlands) in combination with an integrated 16ch posterior coil.

In-vivo imaging was performed in eight healthy volunteers with no reported respiratory disorders, with written informed consent obtained prior to examination. A single coronal slice, centered at the bifurcation of the trachea was acquired in each volunteer. Said slice was measured three times for each of the two trajectories (SPR and radial UTE): once in the inhaled state (breath-hold), once in the exhaled state (breath-hold) - each for a duration of 2.4 s for SPR and 1.99 s for radial UTE - and once continuously during free breathing for a total scan duration of 113 s for SPR and 93 s for radial UTE. TE was kept the same for SPR and radial UTE, whereas TR was kept minimal at 3.1 s for SPR and 2.6 m for radial UTE. The number of interleaves for the breath-hold images was chosen equal to the number of interleaves required to Nyquist sample k-space based on the radial UTE sampling scheme (703 interleaves). All relevant scan parameters are shown in Table 1.

After data acquisition, raw data were exported and processed with MATLAB (MathWorks, Natick, Massachusetts, United States). Images were obtained using gridding [22] with an oversampling factor of 1.25, a Kaisser Bessel kernel for interpolation and in combination with a 2D Voronoi tessellation for density compensation. During trajectory calculation gradient delays were carefully considered [23] and remaining eddy current induced distortions modeled by a mono-exponential decay function [24] with time constant of τ = 43 μ s. No post-processing was applied to any of the presented images.

For evaluation of the trajectory performance for different temporal resolutions when using self-gating, 4 different retrospective self-gating techniques were compared: k 0 -based self-gating ( SG ksp ) , two image-based approaches ( SG img ) , as well as the non-uniform self-gating approach (nuSG).

The k-space based approach [17–19] uses the during each read-out intrinsically measured DC-signal from manually selected coil elements. The DC-amplitudes are bandpass filtered using cut-off frequencies that are individually determined for each volunteer based on a power spectrum analysis. Principal component analysis yields the final respiratory navigator signal that is subsequently sectioned equidistantly along the time axis to obtain the respiratory stages.

The images-based approaches [7, 20], as well as the nuSG approach [21] rely on low-spatial high-temporal resolution images. The required dynamic data is reconstructed applying a sliding window technique with temporal resolution per image of 100, 75 and 50 ms, resulting to 32/39 read-outs for a resolution of 100 ms, 24/29 read-outs for a resolution of 75 ms and 16/19 read-outs for a resolution of 50 ms for SPR/radial UTE, respectively, for the sliding-window reconstruction.

In case of the image-based gating reconstruction, a line was manually placed over the lung-liver interface (LLI). Its line profile automatically underwent gradient analysis to obtain the self-gating signal. The identification of the respiratory stages followed a histogram-based binning approach for the first image-based self-gating technique ( SG img, hist ) and an absolute value based approach for the second image-based self-gating technique ( SG img, abs ) .

The nuSG reconstruction is based on the reconstruction process that is described in detail in [21]. The low-spatial high-temporal resolution images obtained by the sliding-window reconstruction were used to identify a region of interest (ROI) covering the LLI and, by pairwise correlation of all frames, a correlation matrix M c was calculated. Paths of high correlation coefficients were identified using a contour fit approach. Certain motion states are reconstructed from the intersections of the motion state and all grown contour fits combined.

All self-gating reconstruction techniques are described in further detail in [25].

For further assessment of all images, the parenchyma was segmented semi-automatically by manually drawing approximate contours around the left and right lung and by subsequently applying intensity thresholds to exclude large vessels.

The signal-to-noise ratio (SNR) of the lung parenchyma was calculated according to [26] by SNR = 2 − π 2 SI ROI σ BG , (2) where SI ROI is the mean intensity within a selected ROI and σ BG is the noise standard deviation of the pixel intensities within a selected background ROI. The square-root pre-factor in Equation 2 corrects for the Rician distribution of noise in the evaluated magnitude images.

To check for statistically significant differences in SNR, a Wilcoxon signed-rank test for paired samples was performed, with p -values smaller than 0.05 considered significant.

Image sharpness assessment was done across an intensity line profile over the LLI. The position of the pixels corresponding to 25 and 75 % of the maximum signal intensity was identified by a standard metric [21, 27] and the resulting distance used as a sharpness measure.

Data acquisition was completed successfully for six volunteers, the data of two volunteers needed to be excluded due to insufficient image quality of the radial UTE images. Images of one volunteer are exemplarily shown in Figure 1 for breath-hold, as well as all self-gating reconstruction techniques at all investigated temporal resolutions for radial UTE and SPR trajectory. Additionally, Figure 2 shows the images reconstructed using SG img,hist at a temporal resolution of 50 ms for SPR and radial UTE at different windowing, to allow better appreciation of the lung structure. The overall image quality is comparable for all reconstructed images reconstructed at 100 ms temporal resolution. At lower temporal resolutions, especially radial UTE combined with nuSG reconstruction leads to more apparent artifacts.

As they form the core for image-based self-gating reconstruction techniques, as well as nuSG, the low-spatial high-temporal resolution images were further analyzed. A single frame of the sliding window reconstruction for SPR and radial UTE at all reconstructed temporal resolutions is shown in Figure 3. The aliasing artifacts resulting from the undersampling during sliding window reconstruction visibly increases for lower temporal resolutions, but, while radial UTE produces distinct spoke-like artifacts, the noise in the SPR images appears more evenly distributed. Additionally, since with SPR, the entire petal is used for the sliding window reconstruction, the signal intensity is visibly higher in the liver compared to radial UTE. The full sliding window reconstructions can be found in the Supplementary Material S1–S6.

Figure 4 shows the time course of the profile identified over the LLI that is used for image based gating and nuSG for all temporal resolutions for SPR and radial UTE. The self gating signal that was obtained using gradient analysis is shown in red. The algorithm reliably tracks the LLI for 100 and 75 ms for the SPR trajectory with few outliers even at 50 ms temporal resolution. For radial UTE, the algorithm produces several outliers already at 100 ms temporal resolution. The number of outliers increases as temporal resolution decreases, leading to only a low level of correlation between the automatically tracked LLI and the actual breathing pattern. The fitted correlation matrices used for nuSG reconstruction for SPR and radial UTE for all temporal resolutions can be found in the Supplementary Material S7.

The line profiles across the LLI for both trajectories at all temporal resolutions that were used for sharpness analysis of one volunteer are shown in Figure 5 for SPR and radial UTE during breath-hold, as well as for all analyzed self-gating techniques at temporal resolutions of 100, 75 and 50 ms. Especially for radial UTE combined with nuSG, a clear decay in image sharpness for decreased temporal resolution is visible in the line profile over the LLI.

Mean image sharpness is listed in Table 2. The percentage decrease when comparing the sharpness value measured for SPR acquired images with the sharpness obtained from radial UTE acquired images is shown in Figure 6. The data is presented in green for end-inspiration stage and in blue for end-expiration stage. Especially the image-based self-gating techniques yield best results for a temporal resolution of 75 ms. Also noticeable is the approximation of the performance comparing SPR to radial UTE during end-expiration stage when using nuSG.

Mean SNR values in the lung parenchyma inclunding p -values indicating the significance of increase in SNR when comparing SPR to radial UTE can be found in Table 3. When comparing the data acquired using SPR to data acquired using radial UTE, the BH images show no significant increase in SNR, for a temporal resolution of 100 and 75 ms, all gating techniques except for nuSG yield a significant increase in SNR, while for a temporal resolution of 50 ms all gating techniques except for nuSG show significant increase in SNR only for the end-inspiratory stage. The percentage increase in SNR between SPR and radial UTE acquired images is shown in Figure 7. The data is presented in green for end-inspiratory stage and in blue for end-expiratory stage. SNR is higher for SPR compared to radial UTE for all temporal resolutions of all self-gating techniques, as well as for breath-hold images. The highest increase is found at a temporal resolution of 75 ms.