In order to achieve respiratory self–gating (SG) in lung imaging analyzing the changes of the k-space center signal (DC-SG) has been proposed as well as image-based SG (Img-SG), where low spatial / high temporal resolution images are used for deriving the gating signal e.g. by following the lung-liver interface.

We investigated the feasibility to apply DC-SG and Img-SG to UTE 3D lung acquisition of freely breathing rats and calculated lung volume changes, sharpness, signal to noise (SNR) and normalized signal intensity (NSI) in resulting gated images.

6 Winstar rats were imaged on a 7T small animal system (Biospec 7/16, Bruker, Ettlingen, Germany), using a thorax-optimized 4 Rx phased-array coil (Rapid Biomedical, Rimpar, Germany). UTE 3D dataset with a Quasi-Random acquisition scheme [1] were acquired (parameters: TE 8 μs, TR 2.4 ms, isotropic FOV 7x7x7 cm³ , FA 3°, bandwidth 300 kHz, matrix 200x200x200, 6-fold oversampling, acquisition time 30 minutes).

A DC- SG signal was extracted from the absolute value of the k-space center and filtered around the respiratory frequencies (Butterworth filter between 0.6-1.2 Hz or 0.8-1.6 Hz).

Img-SG was extracted from low resolution time-resolved 3D images, reconstructed with a sliding window protocol (temporal resolution 288 ms) with a 3D GRASP algorithm [2]. To extract the SG signal from the images, the magnitude values over time are extracted for each pixel, the respective periodogram of all signals calculated and smoothed with a median filter. The signal with the highest peak between 0.6 and 1.6 Hz was chosen and filtered as the DC-SG.

The SG signals were used for separate data into 10 bins, corresponding to different respiratory stages (Stage 1 to Stage 10 from the highest diaphragm position to the lowest). Moreover, high definition images were reconstructed from the inspiration stage (50% data, i.e. Stage 1 to 5), and were compared in terms of sharpness with ungated images [3].

The lung in the SG images were automatically segmented to calculate lung volumes. SNR and NSI were calculated in manually defined ROIs in lung parenchyma and muscle from 3 slices (SNR: the mean signal in the parenchyma divided by the standard deviation of the signal in the muscle; NSI : mean signal in the parenchyma divided by the mean signal in the muscle).

The statistical significance of the differences in lung volume, SNR and NSI was evaluated applying an ANOVA test for repeated measures with Bonferroni’s correction. Differences of the sharpness index were evaluated with a two-sided paired t-test. Significance level was fixed at 0.05.

Figure 1 shows the resulting image quality of the 3D sliding window reconstruction with and without compressed sensing reconstruction: only with the iterative CS method sufficient image fidelity can be archived for the proposed self-gating method.

Figure 2 shows resulting gated images: with Img-SG clear differences in diaphragm position between different respiratory phases as well as through-plane motion can be appreciated, while the DC-based reconstruction failed to provide functional data.

The extracted lung volumes significantly differ among different respiratory stages (p<0.0001), with decreasing values from Stage 1 to Stage10. The tidal volume resulted to 1.90±0.33 ml. SNR in lung parenchyma was lowest during inspiration (stage 10) with decreasing values towards expiration. Similar, the NSI was lowest during inspiration, but reaching a plateau at Stage 5.

Figure 3 shows a comparison of the resulting image quality between expiratory gated and ungated images. The increase in sharpness can be well appreciated with significant (p<0.001) improved sharpness index of 0.0073±0.002 (0.0053±0.001, ungated).

The presented Img-SG method can be successfully applied for reconstruction of 3D lung images in different respiratory stages. The Quasi Random acquisition scheme in combination 3D GRASP provided sufficient image quality for retrieving the respiratory position from a 3D sliding window reconstruction. The changes in lung volume confirms the validity of the SG method, being in line with previous reported values. In this study, DC-SG compares unfavorably due to only weak changes of the k-space center signal, which may be attributed to minimal changes of the anatomy, in case of volumetric acquisitions. The suggested SG algorithm is straightforward and do not requiring user interaction.

High-definition images in expiration stage resulted significantly sharper than ungated images. SNR and NSI change between respiratory stages, reflecting the change in lung parenchymal density over the respiratory cycle. This changes could be exploited to extract functional information on local ventilation as already been reported for 2D UTE [4], with the advantages of better SNR and isotropic resolution offered by 3D UTE.