INTRODUCTION: Pancreatic ductal adenocarcinoma (PDA) is a leading cause of cancer death with a 5-yr survival rate of only 8%. PDA is characterized by a dense extracellular matrix, which presents a microenvironment that is immune-suppressive and poses a substantial barrier to drug penetration. Genetically engineered mouse (GEM) models of PDA recapitulate the location, biology and genetics of human disease including the dense stroma ¹. Therefore they are useful for testing stroma-directed drugs. Quantitative imaging markers derived from dynamic contrast-enhanced (DCE)-MRI are sensitive to changes in the tumor microenvironment, thus providing a strong rationale to test their utility in the co-clinical trial setting of stroma-directed therapy. However, the rapid heart and respiration rates as well as small blood vessel size in mice pose unique challenges for implementing DCE-MRI method in mice, particularly for direct observation of the arterial input function (AIF). Located in the upper abdomen, pancreatic cancer is highly motion susceptible. Here we present a DCE-MRI technique that achieves motion-robustness simultaneously with high temporal and spatial resolution for capturing the AIF and mapping heterogeneous permeability /perfusion of PDA tumor.
METHODS: A genetic engineered mouse (GEM) that carries Kras and p53 mutation in pancreas specific Cre allele, referred to as KPC model of PDA, was used in this study ². MRI studies were performed on a 9.4T DirectDrive® system (Agilent) interfaced with a 12-cm gradient coil. ECG was monitored (SA Inc) and the core temperature was maintained at 37±0.2°C throughout MRI exams. Longitudinal relaxation time (T₁₀) data were acquired using Look-Locker method we described previously ^{3, 4}. DCE-MRI data were acquired using multi-slice 2D cardiac gated saturation recovery GRE employing golden-angle (111.246°) radial k-space sampling (TR/TE= 2x cardiac period ~ 200/0.91 msec, BW=100KHz, FOV=38 x 38 mm²). Multiple slice groups were planned with one cardiac short-axis slice and contiguous axial slices spanning the tumor. Radial DCE data were regridded to 256 x 256 matrix with or without employing k-space-weighted image contrast (KWIC) filter ^{5, 6} (Fig 1A). For KWIC, 25 views encoded the central k-space region, while 200 views encoded the outer-most regions. Non-KWIC reconstruction utilized 200 views (according to Nyquist criterion) throughout the entire k-space. Because k-space center is over-sampled in radial acquisitions, KWIC allows using fewer spokes at k-space center, thus effectively increasing temporal resolution with sliding window reconstruction ⁶. The AIF was extracted from the blood signal in heart left ventricle ^{3, 4, 7}. After manual definition of tumor ROIs, DCE series, T₁ map and individual-AIF were subject to pixel-wise least-squares fit to the “shutter-speed” pharmacokinetic model yielding parameter maps ^{3, 4}.
RESULTS: The AIF was extracted from regridded images without and with KWIC reconstruction (Fig 1 B-C and D-E). The KWIC reconstruction provide sufficient temporal resolution to fully characterize the initial rise of the AIF. In contrast, the AIF produced from fully sampled images exhibits a blunted (smoothed) profile (Fig 1 F-G). KWIC reconstruction resulted in motion-free high resolution abdominal images of the PDA tumor (Fig 2A-C) despite the absence of respiration-gating. A comparison of selected images from the DCE series reconstructed with full sampling and KWIC demonstrated minimal differences in image quality between the two reconstruction methods (Fig 3A₁-A₅, 3C₁-C₅). The DCE signal time course for the two data sets over the tumor ROI are quite similar while the KWIC time course captures more time points during wash-in of the contrast agent (Fig 3B, D). Tumor T₁, K^trans and k_ep values (T₁ = 2.5 ± 0.15 s, K^trans =0.09-0.33 min^-1, and k_ep = 0.5-1.6 min^-1) obtained from parameter maps (Fig 3E, F, G) are similar to previously reported values ⁴ generated using Cartesian sampling methods at significantly lower spatial resolution.
DISCUSSION: For tumors in motion-susceptible locations such as pancreatic, liver, lung and breast cancer, motion-robustness is crucial for quantitative mapping of DCE-MRI parameters. By leveraging the intrinsic motion resistant properties of radial sampling with view-sharing (sliding window) features of KWIC filtering, we have obtained DCE images of murine pancreatic cancer free of respiratory motion artifacts simultaneously with observation of the AIF with high temporal resolution. The resulting DCE parameter values are consistent with what we have achieved in human DCE-MRI studies on clinical scanners ^8-10.
CONCLUSION: GEM models of human cancer are being used more frequently in preclinical imaging studies. Quantitative DCE-MRI observation of the orthotopic tumors provided by these models is often challenging due to respiratory motion artifacts. Here we have demonstrated that Golden angle radial sampling combined with view sharing techniques can yield artifact free images with high temporal and spatial resolution in pre-clinical DCE-MRI studies.

Acknowledgements

This study was partially supported by U24CA231858 (Penn Quantitative Imaging Resource for Pancreatic Cancer), R21CA198563 and R01CA211337.