Synopsis

DWI of mice with endogenous PDAC tumours were acquired in a 7T MRI system without breath gating.

Inclusion of DWI with b-values above 800 s/mm² in fits substantially improves the qualitative appearance and reduces variance of uniform region fit ADC.

Tumours were grouped (based on histology) by cellularity (amounts of neoplastic cells and stroma, and clustering) and separately by grade. ADC reliably distinguishes tumours of different cellularity (PDAC^low 1.58±0.08; PDAC^med 1.35±0.07; PDAC^high 1.17 ± 0.11; P<0.0001). Grades G2 and G3 were not distinguishable via ADC (1.43±0.15 vs. 1.43±0.16 10^-3 mm²/s), however G4 had significantly lower ADC (1.16±0.10 10^-3 mm²/s, P<0.0001).

Introduction

Profiling of tumour composition of pancreatic ductal adenocarcinoma (PDAC) can be crucial for guiding treatment¹. We have previously demonstrated² that high tumour cellularity (amounts of neoplastic cells and stroma, and arrangement into clusters) and regionally low apparent diffusion coefficient (ADC) are important prognostic biomarkers in human PDAC. Also, we established a classification by cellularity in human and endogenous mouse PDAC tumours using diffusion-weighted MRI (DWI) with a 1.5 T clinical system². The goal of this study was to establish DWI with a 7 T preclinical scanner in the previously-characterized endogenous mouse model of PDAC.

Methods

Tumour model: 36 mice (Ptf1a^wt/cre;Kras^wt/G12D;p53^fl/fl) with 79 PDAC tumour regions

MRI System: Small animal 7T magnet (Agilent/GE) MR901 with Bruker AVANCE III HD electronics and a 31 mm inner diameter volume coil (RAPID Biomedical)

MR Imaging: Tumours were identified with T2w RARE anatomical imaging. DWI used a single-shot EPI-readout with 12 b-values (12-1528 s/mm²), effective TE 34 ms, TR 5000 ms, 10 repetitions, no averaging, 9 slices, 1 mm slice thickness, receive bandwidth 100 kHz, 28x18 mm² FOV, voxel size 0.25x0.25x1 mm³, partial-Fourier factor 1.6, and total scan time 10 min. Imaging was conducted without breath gating.

ADC Fitting: Magnitude images were fit in MatLab with a least-squares model at each voxel for the apparent diffusion coefficient (ADC). To compensate for motion artifacts, which are inconsistent with normally distributed noise, fitting was iterative: First, all repetitions of all b-values were fit. A second fit excluded any repetition further than a threshold (the median of all voxel values in the diffusion-weighted images, including all repetitions and b-values) from the initial fit at that b-value. A third fit repeated this using the result of the second fit to determine excluded points.

Histology: Tumours were fixed in formalin and stained with hematoxylin and eosin (H&E), and slices corregistered to ADC maps. Tumour regions were classified by cellularity² (PDAC^low, PDAC^med, PDAC^high) and grading (G1-G4).

Data Processing: Tumour ADCs were averaged within tumour regions of interest (ROIs) drawn on 3 adjacent slices, and the mean of the slice-means was taken as the average ADC for each tumour. Group mean ADCs were compared pairwise, correcting for multiple comparisons. The importance of high b-values to generating usable ADC maps was assessed by fitting 8 mouse images with and without the four highest acquired b-values (822 to 1528 s/mm²). The standard deviations of voxel ADCs within muscle ROIs (taken as a region with relatively uniform diffusion characteristics) were compared with a ratio paired T-test. GraphPad Prism was used for statistical testing.

Results

An example anatomical and DWI of an endogenous PDAC tumour in a mouse is shown Fig. 1.

The impact of including high b-values on ADC map image quality is illustrated in Fig. 2. The ADC map without high b-values is qualitatively much noisier. Within ROIs drawn on muscle in 8 images with with and without b-values above 800 s/mm², the standard deviations of the ADC values were significantly higher when fit without high b-values: geometric mean of ratios (without- over with-high-b-values) 2.20 (P<0.0001).

Example PDAC tumour H&E histology images are shown in Fig. 3, illustrating tumour cellularity groups.

Tumour cellularity groups had statistically significantly different ADC: PDAC^low 1.58±0.08; PDAC^med 1.35±0.07; and PDAC^high 1.17 ± 0.11; (P<0.0001). G1 tumours were not found in the analyzed cohort. G2 and G3 tumours had very similar ADCs: 1.43±0.15 vs. 1.43±0.16 10^-3 mm²/s. Both had statistically significantly different ADC from G4 tumours: 1.16±0.10 10^-3 mm²/s (P<0.0001). These results are illustrated in Fig. 4.

Discussion

Endogenously grown PDAC tumours in mice were reliably measured with the established free-breathing DWI sequence at 7T.

Inclusion of DWI with b-values above 800 s/mm² substantially improved the qualitative appearance (Fig. 2) and reduced variance in uniform region fit ADCs. This observation may warrant further investigation in clinical scanners.

Separation between groups by cellularity and tumour grade was similar to that achieved at 1.5 T. However, at 7T, resolution was substantially improved, without problematic increase in artifacts or distortion of images, allowing more robust detection and characterization of smaller tumours and better delineation of tumour heterogeneity.

Conclusion

We have established DWI in a 7T preclinical MRI for free-breathing abdominal imaging of mice with endogenous PDAC tumours. Inclusion of high b-values (>800 s/mm²) is necessary for reliable fitting of ADC maps.

ADC at 7T reliably distinguishes endogenous PDAC tumours of different cellularity, and will be used for future preclinical image-guided therapy trials.

Acknowledgements

References

1. Trajkovic-Arsic M., Heid I, Steiger K, et al. Apparent Diffusion Coefficient (ADC) predicts therapy response in pancreatic ductal adenocarcinoma. Sci. Rep. 2017; 7(1): 17038

2. Heid I, Steiger K, Trajkovic-Arsic M, et al. Co-clinical Assessment of Tumor Cellularity in Pancreatic Cancer. Clin Cancer Res. 2017; 23(6):1461-1470