Diffusion MRI is undergoing a striking revival in the application to cancer, and many publications have revealed the usefulness of DWI image in the diagnosis and treatment monitoring of cancer (1), including hepatocellular carcinoma (HCC) (2). Furthermore, recent improvements in MRI scanners have allowed to scan with shorter diffusion times, an interesting approach to disentangle the contribution of different tissue components to the overall diffusion MRI signal (3). Diffusion hindrance is supposed to increase with the diffusion time, as more water molecules hit obstacles, such as cell membranes, the density of which increases in cancer tissues. Thus, our aim was to investigate the relationship between diffusion time and diffusion parameters obtained from 7T MRI using a HCC xenograft mouse model.

Human HCC cell line HepG2 cells (1x10⁶) were injected to the hind limbs of 4 Balb/c nu/nu mice. They developed tumors in 6 weeks with the diameter of 8.8-12.4 mm, and they were imaged on a 7T MRI scanner (Bruker, Germany) using a 1H quadrature transmit/receive volume coil. The SE-EPI acquisition parameters were set as follows; Resolution 250 x 250μm², matrix size 100 x 100, field of view 25 x 25mm² , slice thickness 1.5mm, TE=46.9ms, TR=2500ms, 8 averages, 4 segments. DWI MRI images were acquired using 2 different diffusion times (diffusion gradient duration(δ): 7.2ms, and diffusion gradient separation(Δ): 12ms and 30ms, resulting in the effective diffusion time: 9.6 and 27.6ms) and 19 b values (from 7 to 4105 sec/mm²). The acquisition time for each b value was 80 seconds, and the total acquisition time was 50 min 40 sec. Data analysis was performed using a code developed in Matlab (Mathworks, Natick, MA). ROIs were drawn according to the contrast patterns observed on anatomical and DWI images. Diffusion parameters were retrieved for each ROI. Signals acquired for each diffusion time at b>500 s/mm² to remove IVIM effects was fitted using non-Gaussian diffusion kurtosis model (4):

S(b)=[S0²{ exp [-bADCo+(bADCo)²K/6]}²+NCF]1/2 [1]

where NCF (noise correction factor) a parameter which characterizes the “intrinsic” non-Gaussian noise contribution within the images (4).

A composite, synthetic ADC was also calculated as:

sADC = ln [S(Lb)/S(Hb)]/(Hb-Lb) [2]

where Lb is “low key b value”, Hb is “high key b value” optimized to get the highest overall sensitivity to ADCo and K (5). For this study the Low and High key b values were 438 and 2584s/mm², respectively.

Overall, tumors were highly heterogeneous (Fig.1) All Examples of the plots of the diffusion signal attenuation at two different diffusion times are shown in Fig. 2.The curvature, indicating that diffusion is not Gaussian regardless of diffusion times, clearly increases with the diffusion time, suggesting increased diffusion hindrance. An increase of K (from 0.44 to 0.78, pvalue=0.07) and a decrease of ADC (from 0.75 to 0.56x10^-3mm²/s, pvalue=0.16) was observed when the diffusion time increased, a general trend also found in the rat brain cortex (3). Furthermore, the difference of sADC values was strinking, decreasing from 0.58 to 0.44x10^-3mm²/s (pvalue<0.01) when the diffusion time increased from 9.6ms to 27.6ms (Fig.3).The decrease of ADCo and the increase of K values with the diffusion time are well in agreement with the hypothesis that diffusion hindrance increases with the diffusion time, as more molecules hit boundaries, such as cell membranes. In other words, those results confirm that the deviation of water diffusion from a Gaussian distribution increases with the diffusion time in the range [9-30ms]. Interestingly, those ADCo and K patterns were shifted to longer diffusion times, compared to a previous report obtained in the rat brain cortex (3), suggesting a high sensitivity of the relationship of the diffusion parameters with the diffusion time to the tissue types. This non-Gaussian behavior was very well picked-up using a composite diffusion index, sADC, which is intrinsically sensitive to variations of both ADCo and K (5). The sADC is simple to calculate, requiring signal acquired at only 2 b values, and easy to implement in a clinical setting. Our study also points out the importance of the diffusion time when reporting diffusion parameter values when diffusion is not Gaussian, in order to make results comparable across studies. The investigation of the effects of the diffusion time on non-Gaussian diffusion parameters might provide useful information on tissue types. Non-Gaussian diffusion can be easily characterized using a composite index, such as the sADC. Although our preliminary study requires further validation with a larger cohort and other tumor types, results underline the importance of providing the diffusion times when reporting ADCo and K values.