Introduction

We recently demonstrated a stack-of-spirals ultra-short echo time (UTE) method for high-resolution pulmonary imaging at 0.55T¹. This method used 40% of the data in a stable respiratory phase, resulting in low acquisition efficiency. An iterative motion-corrected (iMoCo) reconstruction framework² can significantly improve the acquisition efficiency of our technique to improve SNR or reduce scan duration. Concomitant fields can be a source of significant image blurring in large field of view spiral imaging at 0.55T and need special consideration^1,3. In this work, we propose a new concomitant field and motion corrected reconstruction (iCoMoCo) for stack-of-spiral pulmonary imaging at 0.55T and present a fast reconstruction which can produce images at scan time. Our technique leverages robust self-gating and trajectory corrections to produce high quality images in both healthy volunteers and patients.

Methods

Data Acquisition: A prototype free-breathing 3D T1-weighted UTE spoiled gradient-echo stack-of-spirals sequence was used for imaging on a 0.55T MRI system (prototype MAGNETOM Aera, Siemens Healthcare, Erlangen, Germany)⁴. Additional readouts were acquired along the superior-inferior (SI) direction to estimate the respiratory signal with temporal resolution of <300ms. All scans used golden-angle increments of spiral interleaves with FOV=45x45x19.6cm³ (LAM: 45x45x26.4cm³), resolution=1.75x1.75x1.75mm³ (LAM: 1.4x1.4x6 mm³), readout-duration=5.4ms (LAM: 3ms), acquisition-time=9.5min, peak-gradient-amplitude=26mT/m (LAM: 33.9mT/m).

Reconstruction: Respiratory waveforms were extracted from the SI-readouts using previously published methods¹. Data was retrospectively sorted into 8 equal bins and image reconstruction was performed in two steps:
Step 1: Respiratory resolved images from each respiratory bin were reconstructed using constrained reconstruction with spatial and temporal total-variation (TV). These images were registered to an end-expiration reference image and the motion fields⁵ (M_k) were estimated for reconstruction.
Step 2: The motion fields were used to perform a motion corrected reconstruction. Simultaneously, concomitant field correction was achieved using multi-frequency interpolation (MFI)⁶; weights and demodulation frequencies were calculated using previously published methods. Figure 1 shows the forward and backward model of the iCoMoCo reconstruction used to solve the following cost function,
$$argmin_X \sum_{n,k}^{N_c,K} ||{W_k \sum_p^P(D_pFC_p)S_nM_kX-d_{n,k}}||_2^2 + \lambda_s TV_s (X)$$
where X is the reconstructed image, d_nk is motion resolved multi-channel data, C_p is MFI weights, M_k is estimated motion field, S_n is coil-sensitivity of the nth channel, F is the non-uniform Fourier transform, W_k is the density compensation weights for each motion state, D_p is the demodulation operator at a given frequency for MFI, and TV_s is the spatial total variation term. N_c represents total number of channels, K is the total number of motion states, and P is the total number of demodulation frequencies.

Data were reconstructed inline using Gadgetron⁷ on a system equipped with Dual AMD EPYC processors, 1TB RAM, 4x Nvidia A100 GPUs. The reconstruction was parallelized over all GPUs to minimize reconstruction time. Density compensation weights were estimated using an iterative method⁸. Motion correction was performed within the Gadgetron framework⁵. Inaccuracies in spiral trajectories were corrected using gradient system impulse response functions^9,10. For comparison images were also reconstructed using our concomitant-field corrected CG-SENSE reconstruction with 40% stable-binned data¹. Apparent SNR (aSNR) was calculated as the ratio of mean signal in the lung parenchyma divided by the standard-deviation of a noise-only region.

Patient Imaging: In this IRB approved study, we demonstrate our method in four healthy volunteers, one patient with lymphangioleiomyomatosis (LAM), a rare cystic lung disease, and one patient with lung nodules. Images were acquired in either coronal or axial views. Data was retrospectively under-sampled to simulate scan-times of 2, 4.5, 7, and 9.5 min. aSNR was calculated for stable-binned CG-SENSE and iCoMoCo reconstruction using data from four healthy volunteers, for each scan-time.

Results

All reconstructions were completed in less than 15 min after the end of scan. Figure 2 shows an example of the respiratory resolved images before and after motion correction to demonstrate robustness of step 1 of the reconstruction, which is critical for iCoMoCo.

Figure 3 qualitatively demonstrates the improvement in SNR with iCoMoCo reconstruction for data acquired in a patient with lung nodules and LAM. Figure 4 demonstrates the improvement in image quality with iCoMoCo reconstruction compared to iMoCo reconstruction for a HV and a patient with LAM. Significant improvement in vessel sharpness and cyst delineation highlight the importance of concomitant field correction at 0.55T.

Figure 5 compares image quality between CG-SENSE and iCoMoCo for scan-times of 2-9.5 min. It shows that iCoMoCo images with scan-time of 4.5 min have comparable image quality to CG-SENSE images acquired in 9.5 min. Figure 5B shows aSNR was measured to be 2.83±1.04, 4.65±0.93, 6.06±0.92, and 7.22±1.20 for iCoMOCO and 1.14±0.42, 2.31±0.48, 3.15±0.53, and 3.84±0.74 for CG-SENSE reconstruction for simulated scan-times of 2, 4.5, 7, and 9.5 min.

Discussion and Conclusion

We have presented a fast inline iCoMoCo reconstruction for high-resolution 3D stack-of-spiral UTE pulmonary imaging on a 0.55T system. We demonstrate that a combination of robust respiratory binning, trajectory corrections, concomitant-field correction and motion correction are necessary for efficient and diagnostic-quality pulmonary imaging at lower field strengths. Using our optimized method, we were able to produce high-quality structural lung images during scan time in less than 15 min. iCoMoCo can improve pulmonary imaging at 0.55T by improving aSNR for a given scan time and by enabling shorter scan times for improved patient comfort.

Acknowledgements

The authors thank Christine Mancini, Kendall O’Brien and Amanda Potersnak for their imaging expertise and Scott Baute, Margaret (Peg) Lowery, Jennifer Henry, Amelia Nargozian, Patricia Julien-Williams and Amanda Jones for assistance with patient recruitment. The authors would like to acknowledge the assistance of Siemens Healthcare in the modification of the MRI system for operation at 0.55T under an existing cooperative research agreement (CRADA) between NHLBI and Siemens Healthcare. The study was supported by the Intramural Research Program, NIH/NHLBI (Z01-HL006257).