In the present study, a multi-parametric PET/MRI scan protocol was suggested to non-invasively characterize tumor heterogeneity in patients with cervical cancer. The heterogeneity in tumor was explored using voxel wise analysis of MRI measurement of diffusion and perfusion parameters as well as PET measurement of angiogenesis and glucose metabolism. The relationship between this comprehensive set of functional data provides potential insight into tumor pathophysiology, which in future can be used for individual therapy adaptation with inhomogeneous radiation dose prescription. This also holds potential for sub-target classification and improved interobserver reproducibility for target volume delineation prior to radiation treatment.
Functional MRI and PET parameter maps appeared heterogeneous within the tumors, and the degree of heterogeneity, as well as the relations between parameters, varied between patients. Tumor heterogeneity is a common feature of the tumor microenvironment and heterogeneity as captured by medical imaging could be an important factor in identifying high-risk volumes as well as prognosis (37, 38).
Tumors were clearly delineated by DW-MRI. The heterogeneity of ADC maps could be attributed to increased diffusion restriction by tumor cells, while cystic and necrotic tissues show higher ADC values. This highlighted the potential of ADC map to contribute to target volume delineation, particularly for tumors with different apparent tissue types as in Figure 1-b. Our findings support the potential of DW-MRI for target volume delineation, as has been described in depth in several studies (17, 31). The use of a b-value of zero may result in overestimation of the ADC due to contributing perfusion effects (14). However, it is commonly used in the setting of cervical cancer imaging (12, 15, 23).
A moderate to high level of FDG uptake was observed across patients. On the other hand, the patients show a very variable level of RGD uptake including two patients had very low RGD uptake. FDG-PET is widely used as a surrogate parameter of cell viability and glucose metabolism (35, 39), while RGD uptake is used to assess integrin αvβ3 expression and angiogenic tumor cells (40). The heterogeneity of both FDG and RGD uptake within the same cancer type could be due to differences in properties such as growth rate, vascularity and necrosis in the tumor cell population (21). In an explorative analysis, we found no relation of RGD and FDG uptake to histological tumor type and tumor grade (Table 1).
Among other frequently used pharmacokinetic models, the standard Tofts model has been reported to be preferable in analysis of clinical DCE-MRI data (41). The parameters estimated using the Tofts model had a high degree of variation within the patient cohort. The volume transfer constant Ktrans is determined by the blood perfusion and the permeability surface area product of the vessel wall in varying proportions and expected to correlate with RGD uptake as both parameters are supposed to represent angiogenesis and active endothelial cells (17, 42). However, we did not observe any significant relation between perfusion metrics and other parameters. This finding is contrary to previous observation by Metz et al. (21), wherein they found a tendency toward higher values of tumor perfusion in areas with more intense RGD uptake. The results are also not supporting the claim that tumor perfusion could be simulated by FDG uptake (21), which may occur when the tumor blood flow is inadequate and requires a well-developed tumor vascular supply.
Except for one patient with a small tumor (volume of 10.5 cm3), the RGD uptake was generally lower than FDG which is in line with previous results for lung cancer (40). The voxel-by-voxel analysis of parameter maps revealed weak correlations between RGD and FDG uptake, which varied strongly within the patient cohort, indicating that they provide different and potentially complementary information about the tumor. Our results demonstrated an average correlation coefficient of -0.7 between FDG uptake and ADC values, in line with earlier studies (26, 43, 44), while the correlation between ADC and RGD uptake was weaker.
On a voxel-wise level, there appear to be an L-shaped relation of FDG uptake and ADC values in the 2D joint histogram (Figure 3-a). This type of relation between ADC and FDG uptake has been widely observed for different regions and cancer types (45, 46). The same trend can be observed between RGD uptake and ADC (Figure 3-b). However, the 2D histogram analysis suggests a cluster of tumor voxels with low ADC and low RGD uptake. The majority of the voxels corresponding to this area belong to patient 2 and patient 5, both having tumors with very low RGD uptake. We went one step further to evaluate the combination of ADC, FDG, and RGD uptake and explore their ability to identify tumor sub-volumes. Distinct clusters in Figure 4, resulting from GMM model were indicative of a viable tumor region (blue dots with high RGD uptake, intermediate FDG uptake, and low ADC value), and a region with mixed necrosis (red dots with low RGD, low FDG, high ADC). The third cluster however is more equivocal with low ADC values, variable FDG uptake and almost no RGD uptake (green dots). The third cluster corresponds to the previously referred to feature in the joint histogram of ADC and RGD uptake (Figure 3-b, lower left). We may speculatively interpret the third cluster as representing solid tumor tissue however with a small number of newly formed vessels. Tissue histology will be needed to shed more light on the possible significance of the cluster analyses. A cluster analysis has been done before in a preclinical study considering FDG and ADC to provide a more stable metric for multi-parametric tumor characterization (28). A similar study was conducted on patients with non-small cell lung cancer (NSCLC) (47). They created 4 clusters based on single value as a threshold to distinguish between high and low ADC value and FDG uptakes.
In studies employing multi-parametric imaging, alignment between different modalities will always be a potential limitation. It highlights the major advantages of combined PET/MRI scanner in the more straightforward voxel-by-voxel analysis of MR and PET parameters. However, it is still a critical issue for studies of the pelvic area, due to changing patient anatomy in particular originating from urinary bladder filling. Depending on the time interval between acquisition of each image as well as the proximity of the tumor to the bladder, it can cause misalignment over tumor volume. Our solution to this issue to manually perform local registration between images could reduce the effect.
It needs to be stated that the current results are based on a low number of patients. The focus of this study, however, was to explore feasibility and the potential of multi-parametric PET/MRI for cervical cancer. Thus, it might contribute to the future design of individually adapted treatment approaches based on multi-parametric functional imaging. Studies with a larger cohort are demanded to further evaluate our findings. Its complementary roles to improve delineation reproducibility should be explored further with histologic validation.