Analysis of pairwise correlations in multi-parametric PET/MR data for biological tumor characterization and treatment individualization strategies.

Abstract

Biological tumor characterization based on functional and molecular imaging might be highly valuable for radiotherapy (RT). On the one hand, it could allow for an improved target volume definition and an individualized dose prescription within the tumor according to local biological characteristics. Such dose painting strategies can be readily applied with the technical availability of intensity modulated RT (IMRT). Moreover, functional imaging might be of high value for early response assessment and potential treatment adaptation in the course of fractionated RT [1, 2]. Other fields of application are the assessment of chemotherapy and the application of targeted agents, such as hypoxia-sensitizing or antiangiogenic drugs [3, 4]. Both positron emission tomography (PET) and magnetic resonance imaging (MRI) may provide functional information beneficial for personalized treatment strategies. PET imaging using [ 18F]-fluorodeoxyglucose (FDG) can be used to monitor glucose metabolism, whereas the hypoxic status of the tumor can be assessed using dedicated tracers such as [ 18F]-fluoromisonidazole (FMISO). Diffusion weighted MRI (DW-MRI) provides the possibility to quantify the diffusion of water molecules, which is related to cellular density [5]. Dynamic contrast-enhanced MRI (DCE-MRI) yields a temporally varying signal due to the distribution of contrast agent in blood pool and tissue. By compartmental modeling estimates of quantitative physiological parameters can be derived [6]. With the advent of combined PET/MR imaging [7, 8] the acquisition of simultaneous, intrinsically registered PET and MR data has become possible. This facilitates the comparison and combined analysis of PET- and MR-derived functional imaging data. Simultaneous PET/MR may thus be of high potential for treatment individualization [9, 10]. Recent studies have associated different functional imaging information with RT outcome for head and neck (HN) cancer. This applies to FDG-PET [11, 12], static as well as dynamic FMISO-PET [13–16], apparent diffusion coefficients (ADCs) inferred by DW-MRI [17], as well as DCE-MRI [18, 19]. These studies provide a rationale to adapt RT treatment plans according to functional imaging information. It is not clear yet if datasets from different functional imaging modalities are completely complementary, or if information is to some extent redundant. Initial analyses of correlations between different functional datasets have already been performed in recent studies. The studies of Rajendran et al. [20] and Thorwarth et al. [21] revealed good voxel-by-voxel correlation of FDG and FMISO in some HN tumors, whereas others showed no clear correlation. The biological basis of the observed correlations may be the hypoxia-inducible factor 1 α (HIF 1 α) [20]. Similar results were obtained by Zegers et al. [22] comparing uptake of FDG and the hypoxia PET tracer [ 18F]-HX4 in patients with non–small cell lung cancer. Houweling et al. [23] quantified correlations between FDG and ADC maps of HN tumors on a voxel level, and found a negative correlation in most patients. Both Newbold et al. [24] and Donaldson et al. [25] found correlations between hypoxia derived from pimonidazole staining and DCE-derived parameter maps on a region-of-interest (ROI) level. A study by Jansen et al. [26] found that neck nodal metastases with positive FMISO uptake differed significantly in median Ktrans values from those with no FMISO uptake. Earlier studies have shown that a dynamic imaging protocol may be superior compared to a single time frame for hypoxia quantification using FMISO-PET [16]. However, in addition to a late static scan several hours post injection (p.i.), such a dynamic protocol requires a PET acquisition during tracer wash-in in the first minutes p.i. [27], which may hamper its usage in clinical routine. A positive correlation result between early FMISO and DCE information would potentially provide the possibility to infer early FMISO information from DCE, which would facilitate its clinical usage. To address the question if available functional information of PET/MR is complementary or to some extend redundant, this study extends beyond existing studies by considering a comprehensive set of functional data. Correlations of various functional datasets are quantified on a voxel as well as on a regional level within HN tumors by means of the Spearman correlation coefficient. For the analysis, FDG-PET, FMISO-PET acquired in the wash-in, as well as in the retention phase, ADC maps extracted from DW-MRI, and DCE-MRI derived maps are taken into account. The study is a first explorative, hypothesis generating approach to investigate the utilization of integrated PET/MR for personalized treatment strategies.