It has been shown repeatedly that breathing oxygen shortens the $T_1$ relaxation time in ventilated regions of the lung$^\text{1}$. The most widespread methods to quantify $T_1$ in the lung is the inversion-recovery SNAPSHOT FLASH sequence$^\text{2}$. However, this method is gradient-echo based and therefore suffers from signal attenuation due to air tissue interfaces. Furthermore, an echo time (TE) dependency of $T_1$ has been demonstrated using an ultra-short TE sequence$^\text{3}$: Avoiding $T_2'$-decay, extra-vascular components contribute to the signal, influencing the average $T_1$, promising additional diagnostic information$^\text{3}$. Since UTE-sequences are time intensive, we propose an inversion-recovery spin echo SNAPSHOP FLASH sequence, employing self-refocusing excitation pulses in the regime of weak dephasing, which allow an echo time - measured from the end of the RF-pulse - that exceeds the pulse duration$^\text{4}$. It is shown that quantitative $T_1$ and proton density (PD) maps of the entire lung can be acquired within $5.5~s$. The highly undersampled data is addressed in an iterative reconstruction.

Data were acquired in a healthy volunteer during a breath-hold at full expiration. A 3T PRISMA scanner (Siemens, Erlangen, Germany) was used equipped with the manufacturer's spine- and body-array of which 30 channels were used. All experiments were performed with the subject breathing room air and were repeated under the consumption of oxygen. An inversion-recovery SNAPSHOT-FLASH sequence was modified to use a spin echo generating excitation pulse$^\text{4}$. The employed pulse has a duration of $150~\mu\text{s}$ and an echo time of $630~\mu\text{s}$ measured from the end of the RF-pulse. The pulse shape is depicted in Fig.1 along with the corresponding phase evolution of the excited spin isochromats. A flip angle of $1.2^\circ$ was used. For comparison, a $100~\mu\text{s}$ rectangular pulse was used.

The nominal spatial resolution was $5~\text{mm}\times5~\text{mm}\times10~\text{mm}$ with a $FOV=640~\text{mm}\times640~\text{mm}\times240~\text{mm}$ in coronal orientation. With $TR=1.6~\text{ms}$ the total acquisition time was $5.5~\text{s}$. Spatial encoding was performed with a stack-of-stars trajectory: While sweeping through $k_z$, an increment of $\alpha_\text{gold}\cdot{N_{shots}}/N_z$ was used, where $\alpha_\text{gold}\approx111^\circ$ is the golden angle, $N_{shots}$ is the total number of excitations and $N_z$ is the matrix size in z-direction. After acquiring one spoke for each $k_z$, the whole trajectory is repeated rotated by $\alpha_\text{gold}$, ensuring good k-space coverage both within each partition as well as in comparison to the neighboring partitions.

In total 3456 spokes were acquired, which corresponds approximately to 72% of Nyquist sampling. With this data, $T=144$ images were reconstructed, using one spoke per $k_z$-partition and minimizing the following cost function: $$\tilde{x}=\arg\min_{x~\in~\mathbb{C}^{N \times T}}\sum_{t=0}^{T-1}\left|\left|\mathbf{E}_{t,:,:}\cdot{x_{:,t}}–S_{:,t}\right|\right|_2^2+\lambda^2\sum_{n=0}^{N-1}\left|\left|x_{n,:}-\mathbf{P}_{\mathcal{A}}(x_{n,:})\right|\right|_2^2\;.$$ The search variable $x$ contains all $N$ voxels of all $T$ time-frames. The first summand is the data consistency term with the forward operator $\mathbf{E}$ which incorporates a non-uniform FFT and ESPIRiT$^\text{5}$ coil sensitivities. The second term compares the time series of each voxel with its projection $\mathbf{P}$ onto the manifold $\mathcal{A}=\{\exp(i\phi)(a-b\exp(-t/c))\;|\;\phi,a,b,c\in\mathbb{R}\}$. Because this algorithm approaches the Bloch response recovery via iterated projection$^\text{6}$ algorithm (BLIP) when setting the regularization parameter to $\lambda\rightarrow\infty$ we refer to it as generalized BLIP (gBLIP). Given this projection, $T_1=c(b/a-1)$ reveals the desired relaxation time for Look-Locker sequences. For computational efficiency, the projection was performed with a precomputed dictionary.

Fig.2 shows one slice of the resulting PD-maps. One can observe an increase of the parenchyma signal when employing the proposed spin echo pulse (b) compared to the traditional gradient-echo IR-SNAPSHOT-FLASH (a). In a region of interest in the upper left lung, the signal increase was measured to be 38%. Furthermore, one can observe an increase of the fat signal, which is a side effect resulting from fat-water-shift in combination with the excitation profile of the RF-pulse$^\text{4}$. Fig.3 shows a slice of the $T_1$-maps under the consumption of oxygen. The spin echo image shows a similar reduction of the relaxation time compared to UTE-sequences$^\text{3}$. Smoothed maps of the $\Delta{T_1}=T_1^{\text{air}}-T_1^{\text{O}_2}$ are shown in Fig.4, overlayed onto PD-maps. $\Delta{T_1}$ is slightly reduced when employing the SE-pulses and the structure of $\Delta{T_1}$ is changed.

In this study, a similar increase of the parenchyma signal has been observed compared to previous studies$^\text{4}$. Furthermore, the reduction of $T_1$ due to signal contributions of the extra-vascular components is strongly comparable to the reduction observed with UTE-sequences$^\text{3}$. Thus, our experiment supports the hypothesis of $^\text{3}$ that the change in $T_1$ results from extra-vascular signal contributions with short $T_1$ that are less visible in gradient-echo images, since their neighborhood to air results in a fast $T_2'$-relaxation. Overall, the results demonstrate the feasibility of the proposed IR-SE-SNAPSHOT-FLASH sequence for quantitative lung imaging with additional diagnostic information. The changes in the $\Delta{T_1}$-maps are not yet fully understood and will be subject of future investigations.