Improved vessel–tissue contrast and image quality in 3D radial sampling‐based 4D‐MRI

Abstract Purpose In radiation treatment planning for thoracic and abdominal tumors, 4D‐MRI has shown promise in respiratory motion characterization with improved soft‐tissue contrast compared to clinical standard, 4D computed tomography (4D‐CT). This study aimed to further improve vessel–tissue contrast and overall image quality in 3D radial sampling‐based 4D‐MRI using a slab‐selective (SS) excitation approach. Methods The technique was implemented in a 3D radial sampling with self‐gating‐based k‐space sorting sequence. The SS excitation approach was compared to a non‐selective (NS) approach in six cancer patients and two healthy volunteers at 3T. Improvements in vessel–tissue contrast ratio (CR) and vessel signal‐to‐noise ratio (SNR) were analyzed in five of the eight subjects. Image quality was visually assessed in all subjects on a 4‐point scale (0: poor; 3: excellent). Tumor (patients) and pancreas (healthy) motion trajectories were compared between the two imaging approaches. Results Compared with NS‐4D‐MRI, SS‐4D‐MRI significantly improved the overall vessel–tissue CR (2.60 ± 3.97 vs. 1.03 ± 1.44, P < 0.05), SNR (63.33 ± 38.45 vs. 35.74 ± 28.59, P < 0.05), and image quality score (2.6 ± 0.5 vs. 1.4 ± 0.5, P = 0.02). Motion trajectories from the two approaches exhibited strong correlation in the superior–inferior (0.96 ± 0.06), but weaker in the anterior–posterior (0.78 ± 0.24) and medial–lateral directions (0.46 ± 0.44). Conclusions The proposed 4D‐MRI with slab‐selectively excited 3D radial sampling allows for improved blood SNR, vessel–tissue CR, and image quality.

Despite wide clinical adaptation, 4D-CT has a few limitations.
First, intrinsically, low soft-tissue contrast makes it difficult to visualize tumors in tissues of similar electron densities. Implanted fiducials, while useful in some cases to ameliorate the problem, are invasive and associated with imaging artifacts that may interfere with outcome assessment. 11,18 Second, 4D-CT images, due to its two-dimensional (2D) acquisition nature and the need for slice resorting, are prone to stitching artifacts that substantially undermine the visualization of tumors and organs. 15,16,19 Third, 4D-CT does not show adequate contrast between tumors and blood vessels. This makes it challenging to target radiation dose to the area of tumor that's in contact with the blood vessels, which is needed in, for example, pancreatic cancer patients to downstage unresectable tumors and obtain margin negative resections for significantly prolonged survival. 2,9 Alternatively, 4D magnetic resonance imaging (MRI) has been used to assess respiratory motion for radiation treatment planning. 1,3,[5][6][7] Compared to CT, MRI provides superior soft-tissue contrast and is free of ionizing radiation. Most of previous 4D-MRI techniques inherit the concept from 4D-CT and are based on multiple 2D acquisitions followed by slice sorting in the image domain. The resultant images are poor in slice resolution and prone to stitching artifacts. [5][6][7] Recently, continuous 3D acquisition with retrospective data sorting techniques in the k-space domain has been proposed to provide stitching artifactfree 4D-MRI images with high spatial resolution. 2,10,12,14 However, for methods involving 3D k-space sorting, undersampling artifacts (e.g., streaking in 3D radial acquisitions) may become evident due to drastic undersampling in patients who have a highly irregular breathing pattern, thus impairing overall image quality. In addition, to the best of our knowledge, current 4D-MRI techniques have not looked into improvements in blood vessel delineation, which is potentially important for radiotherapy planning in pancreatic cancer patients as mentioned above. MR techniques using balanced steady-state free precession (bSSFP) with T2-preparation have shown to improve vessel delineation in coronary MR angiography applications. 20,21 Further, bSSFP methods have also been adopted in 4D-MRI abdominal imaging, however, vessel-tissue contrast appears inadequate and further parameter optimization may be needed. 22 In general, bSSFP and T2-preparation sequences work well in lower magnetic fields, such as 1.5 T.
However, at 3.0 T, these techniques may be susceptible to field inhomogeneity and more prone to image artifacts.
In this study, we aimed to develop an improved vessel-enhanced method at 3.0 T using a slab-selective 3D radial sampling-based gradient recalled echo (GRE) acquisition approach. In addition, improvement in image quality was also explored for the proposed technique.
A pilot study including both healthy subjects and patients was performed to demonstrate these technical improvements.

2.A | Sequence development
In 3D radial sampling-based methods, non-selective (NS) excitation with hard radio-frequency (RF) pulses is typically used to excite a volume considerably larger than the prescribed field of view (FOV).
As a result, blood spins will experience a large amount of RF pulses prior to entering the FOV and thus exhibit a substantially decayed signal level and reduced contrast to stationary tissue spins.
Slab-selective (SS) excitation, in contrast, would ensure fresh blood spins entering the excited volume, thus improving the contrast between blood vessels and stationary tissues. The schematic diagram of NS and SS excitation approaches and simulations of blood and tissue signal and their contrast vs. the blood spins' traveling distance are shown in Fig. 1. In simulations, a general GRE signal equation was used with T1 values of 1500 and 725 ms for the blood and tissue, respectively. The velocity of the blood was assumed as 1 m/s to reflect the scenario in the abdominal aorta. As shown in Fig. 1(b), the blood signal (equivalently represented by the longitudinal magnetization) gradually decreases with the blood spins' traveling distance, whereas the stationary tissue signal remains in a steady state due to repetitive RF excitations; vessel-tissue contrast is well preserved with the in-flow effect by using the SS excitation approach (the blue-dash box). Of note, slower blood flow and tortuousness of the smaller blood vessels may experience more RF excitations, which may lead to a faster signal decay and reduced vessel-tissue contrast than shown in simulations herein.
In addition to potential improvements in vessel-tissue contrast, the SS approach may also help suppress the signals form peripheral structures in the superior-inferior direction. These signals would otherwise contribute to streaking artifacts in the FOV when drastic k-space undersampling occurs.
Hence, an SS-excited 3D radial sampling-based technique was developed in order to achieve the improvements above. The MRI sequence was implemented on the basis of a 4D-MRI framework-3D radial sampling with self-gating-based k-space sorting (SG-KS). 2,23

2.B | Experiments
To evaluate the performance of the SS-excited 3D radial samplingbased 4D-MRI technique (SS-4D-MRI), six patients (mean age: 63.5 AE 17.2 yr; all males; 1 lung, 1 liver, 1 esophagus, and 3 pancreatic) and two healthy subjects (mean age: 36.5 AE 0.7 yr; 1 female) were prospectively recruited with an institutional review board approval and informed consent. Two lesions in the lung patient, three lesions in the liver patient, and a single lesion in all other patients (esophagus and pancreatic) were present.
All imaging studies were performed on a hybrid PET-MRI system with a 3.0-Tesla magnetic strength (Biograph mMR, Siemens Healthcare, Germany) using a body matrix coil and spine coils. The advantages of the SS excitation approach were investigated through a comparison study involving SS-4D-MRI and NS-4D-MRI scans. For a fair comparison, the two scans employed the same SG-KS 4D-MRI framework. 2,13 Specifically, a GRE sequence with 3D radial k-space sampling was used for data acquisition. Self-gating was used for respiratory motion detection and 2D golden-means k-space trajectory was used to enable arbitrary k-space sorting. Acquired k-space data DENG ET AL.

| 251
were sorted into ten respiratory bins and then underwent conjugate gradient sensitivity encoding reconstruction combined with iterative motion correction and averaging. 8

2.C.3 | Evaluation of motion trajectory
To ensure that the motion information is maintained when using SS-  Intraclass correlation was used to determine the agreement in the motion trajectory between the two approaches. In all tests, statistical significance was defined at P < 0.05 and data were presented as means AE standard deviations.

| RESULTS
All subjects successfully underwent SS-4D-MRI and NS-4D-MRI scans. Figure 2 shows the visual comparisons of vessel-tissue con-    11,18 Radiation therapy with simultaneous integrated boost has been proposed to improve tumor resection rate by delivering a greater sterilizing dose to tissues surrounding the vessels but its success clearly depends on how accurate these vessels can be localized and motion characterized. 15,16,19 Due to the low contrast of vessels in both noncontrast 4D-MRI and 4D-CT, this has been a major roadblock for pancreas radiotherapy that requires repeated imaging.

| DISCUSSION
The SS-4D-MRI approach exploits the effect of flow-related enhancement. As fresh blood first enters the imaging volume of interest and experiences fewer RF pulses than stationary tissues, blood signal is markedly higher than that of tissue, creating appreciable vessel-tissue contrast. This phenomenon applies to both arteries and veins as shown in Fig. 2(a). In this study, a 200-mm slab thickness was used as this imaging volume appeared sufficient to cover variously sized tumors in a single organ. Adequate vessel-tissue contrast was observed in the tumor involved vessels, especially near the aorta. Images from SS-4D-MRI showed overall higher vessel-tissue contrast compared to images from NS-4D-MRI. The SS-4D-MRI technique shows the potential of providing such information without the need for repeated administration of intravenous contrast for fractionated radiotherapy.
As an additional benefit of SS excitation, image quality was also improved with evidently reduced streaking artifacts, which is otherwise likely present with drastic k-space undersampling in conventional 3D radial sampling and k-space sorting-based 4D-MRI. By exciting a smaller imaging volume with the SS-4D-MRI approach, the excess signal from the superiorly and inferiorly peripheral structures has less contribution to the reconstructed images, thus, reducing the amount of streaking artifacts normally seen in the NS excitationbased methods. Hence, SS-4D-MRI could potentially afford higher undersampling ratio (shown in the supplemental file) for patients with irregular breathing patterns, who may require high data rejection rates, or permit shorter imaging time for patients with stable breathing. However, severe breathing abnormality was not observed in the small subject group, and thus the potential improvement with the proposed method was not systematically investigated.
The technique has several limitations which could be improved.
The proposed SS-4D-MRI approach is mainly dependent on the inflow effect for vessel enhancement, which leads to a gradual decrease in blood signal as flowing spins traverse the slab. In addition, as blood flow velocity is dependent on vessel size and location, vessel-tissue contrast may differ from large to small vessels within a subject. Furthermore, blood flow velocity may vary from subject to subject, resulting in variable vessel-tissue contrast from subject to subject. These issues could potentially be alleviated by using optimal placement of the selective slab or flexible slab thickness. One can place the selective slab so that the blood vessel of interest is in close proximity to its upstream edge of the slab, thus maximizing the in-flow effect. This is particularly beneficial for slow-flow small vessels. Depending on the imaging volume of interest, flexible slab thickness could also help to achieve desired vessel-tissue contrast.
This study was performed in a small cohort of heterogeneous subjects to show the feasibility of enhancing vessel-tissue contrast.
A study with a larger number of patients with blood vessel-involved pancreatic cancer is needed to clinically evaluate the efficacy of the vessel contrast-boosting approach in treatment outcome for this particular patient population.

| CONCLUSION S
A SS-excited radial sampling-based 4D-MRI technique was developed and tested in healthy volunteers and cancerous patients. The technique significantly improves vessel-tissue contrast and image quality, resulting in a set of respiratory-resolved 3D volumetric MRI images with high isotropic resolution and superior soft-tissue and vessel-tissue contrast.

CONFLI CT OF INTEREST
The authors have no relevant conflicts of interest to disclose.