In vivo liver cancer research commonly uses rodent models of liver tumor growth. There is a lack of reliable endpoints to assess tumor volume changes over time. In particular, histology only provides tumor volume at a single, final, time point. With regard to radiological assessment, studies exist with both negative and positive contrast agents. However, radiological techniques without exogenous contrast agent to quantify liver tumor growth have not been standardized. By using novel image weighting (a combination of 2D T2 and 3D GRE) and 3T clinical MRI, we developed a semi-automated volume quantification method to assess tumor growth in two rat models of HCC.

Rats (n=11) consisted of two distinct groups: Fischer rats (n=5) underwent 30 min of liver ischemia before injecting 2.0x10⁶ syngeneic hepatocellular carcinoma (HCC) in the portal vein; Buffalo rats (n=6) not undergoing ischemia were injected with 2.5x10⁵ syngeneic HCC cells in the portal vein. Fischer rats were scanned at 3 weeks, followed by sacrifice for histology (H&E staining). Buffalo rats were assessed radiologically at 3 and 4 weeks and blood samples were taken at similar time points to measure α-fetoprotein (AFP) level.

MRI used a Siemens MAGNETOM Trio/Prisma 3T clinical scanner using the wrist coil for homogenous liver SNR. MRI included 2D T2 for high contrast and 3D isotropic GRE acquisitions for segmentation. We acquired conventional 2D-TSE with TE/TR/FA=57ms/3500ms/180°, 3 averages, 0.37mm pixel resolution, 1mm slice thickness. For 3D isotropic resolution GRE protocol changes, from 0.35mm TE/TR/FA=0.07ms/9.6ms/10° for Fisher rats to TE/TR/FA=4.5ms/11.8ms/30° for Buffalo rats, were made during data collection due to scanner upgrade.

The use of two MRI signal contrasts allowed separation of tumor regions, normal liver parenchyma, vessels and pockets of oedema. The custom analysis software (PMU, Salzburg) allowed simultaneous segmentation of either of two images (3D and calculated) giving volume quantification. The semi-automatic intensity-based threshold method selects regions under an empirical threshold (100 arbitrary units) around a user-defined point, with a maximum diameter of 20 pixels and edge stop to avoid leakage, with possibility of manual correction.

Histology quantification was used as a gold standard, with quantification of full volume on histology using the same segmentation technique on the 5μm slice (assuming interpolation over the 500μm slice gap). Clinical histological assessment for volume commonly calculates an assumed uniform volume from diameters. Statistical comparison was carried out for a wide range of tumor sizes, by assessing the 15 distinct tumors in 5 Fischer rats by MRI and histological volume and correlation between MRI volume and AFP, a standard clinical blood test, at multiple timepoints for the 6 Buffalo rats.

T2-weighting gives high tumor contrast but is limited to 2D due to time constraints. GRE has poorer distinction between vessels and tumors than T2. The contrast-to-noise ratio (CNR), ((tumor-normal)/sd noise) is on average doubled, from 10 on GRE images to 20 on the T2/GRE fusion (with interpolation to the GRE isotropic slice thickness). Figure 1 shows contrast and image quality of multiple small tumors in the Buffalo model for contrast and image quality.

For Fischer rats, 15 larger tumors were observed in 5 rats and segmented in both MRI and histology (figure 2). The volume segmentation on MRI and histology is shown in figure 2. Volumes compared to histology show good agreement (figure 3) with y=1.24x-8.56 R²=0.84 for the 14 small-medium tumors and y=0.87x+38.07 R²=0.997 including a tumor an order of 10x larger.

In Buffalo rats, we plotted the dynamic MRI volume acquisitions against repeated blood levels of AFP(ng/ml) and found an excellent correlation (y=0.74x+1.14, R²=0.98) as shown in figure 4 with a wide range of tumor volumes.

Combination of the two images provides better both 3D isotropic resolution and contrast between tumors, liver and vessels. Vessel to tumor contrast is lacking in 3D images, but with T2 imaging in the protocol, both contrast and resolution advantages can be exploited. Segmentation enables more accurate assessment in situations such as irregular-shaped or confluent tumor nodules, or very high numbers of small tumors and is significantly faster than manual contouring in such cases. There is good agreement between MRI assessment and histological volume (despite shrinkage on fixing) for endpoint studies or AFP for follow-up.We have developed a 3D volume quantification method that enables follow-up and analysis of complex liver tumor morphology. GRE provided hypointense signal of tumor tissue with high 3D isotropic resolution and resistance to motion artifact. These images could be further improved for segmentation using a semi-automated threshold. Segmented tumor volumes correlated with histological volumes and blood levels of AFP.