ESTRO 38 Abstract book

S1114 ESTRO 38

site Tübingen, and German Cancer Research Center DKFZ, Heidelberg, Germany

Purpose or Objective To investigate principal component analysis (PCA) as an alternative strategy to pharmacokinetic modelling for quantitative and more robust analysis of dynamic PET or contrast-enhanced (DCE) MRI data on a voxel level. Material and Methods In 31 immuno-deficient nude mice, five different ectopic head and neck squamous cell carcinoma xenografts (subcutaneous, hind leg) were investigated with simultaneous PET/MRI (7T, Bruker Biospec). Tumor radiation sensitivity, i.e., tumor control dose 50% (TCD 50 ) to clinically relevant irradiation with 30 fractions in 6 weeks have been previously published [1]. The imaging protocol included dynamic FMISO-PET over 90 min, T2- weighted (T2w) MRI and DCE T1-weighted MRI (750 s, 139 frames). Tumor tissue was delineated on T2w MRI data. Necrotic tissue was excluded based on DCE-MRI information (low temporal median enhancement). The Brix model was fitted to the DCE-MRI tumor data on the voxel level. Standard deviations (SD) obtained from the fit covariance matrix were used to estimate the robustness of the fit parameters. For PCA, DCE signal data S(t) was converted to relative signal increase RSI(t) = (S(t) – S 0 ) / S 0 , where S 0 represents the mean signal before contrast agent injection. PCA was then applied to the RSI data of the total set of tumor voxels of all animals. Finally, parameter maps were calculated for both Brix and PCA. For each map, different thresholds were tested to identify volume fractions that would allow for stratification in terms of TCD 50 . Results DCE-derived Brix parameters presented with highly varying SDs resulting in large fit inaccuracies on the voxel level. A Brix was found to be more robust than k ep . However, 35 ± 14% of the voxels within a tumor (mean ± SD over all tumors) showed a relative SD of A Brix greater than 0.3. Both visual and mathematical examination of voxel curve data indicated that PCA allows for dimensionality and noise reduction in the recorded DCE data. Reconstructing the data with just the first two principle components reduced the number of variables extremely (139 time points) while most of the temporal information was still captured. Further principal components seem to rather depict noise (Figure 1). The percentage of the total variability expressed by the principal components PC1 to PC5 was 94.22, 1.95, 0.18, 0.11 and 0.10%, respectively. In a first analysis, similar results were found for PCA of dynamic FMISO-PET data. Threshold-based analysis of parameter maps indicated that the relative volume presenting with PC2 < -0.12 correlates with TCD 50 (Figure 2). No such correlation was found for Brix parameters, other principal components, or any mean parameter values. Conclusion PCA may be a more robust alternative to pharmacokinetic modelling for the analysis of functional image data on a voxel level and may reveal relevant information e.g. in terms of prediction of tumor response to therapy. Further investigation of PCA on patient data seems promising. [1] Yaromina et al., Radiother Oncol 96:116 (2010)

EP-2030 Multiparametric MRI and FMISO PET in HNSCC and its relation with outcome N. Wiedenmann 1,2 , H. Bunea 1,2 , H. Rischke 1,2 , A. Bunea 1,2 , N.H. Nicolay 1,2 , L. Majerus 1 , L. Bielak 3 , A. Protopopov 3 , U. Ludwig 2,3 , M. Büchert 2,3 , C. Stoykow 2,4 , M. Mix 2,4 , M. Bock 2,3 , A. Grosu 1,2 1 University of Freiburg, Department of Radiation Oncology- Medical Center University of Freiburg, Freiburg, Germany ; 2 German Cancer Research Center DKFZ- German Cancer Consortium DKTK, Partner Site Freiburg, Heidelberg, Germany ; 3 University of Freiburg, Department of Radiology- Medical Physics- Medical Center University of Freiburg, Freiburg, Germany ; 4 University of Freiburg, Department of Nuclear Medicine- Medical Center University of Freiburg, Freiburg, Germany Purpose or Objective Purpose of the current study was to assess the effect of radiochemotherapy (RCT) on tumour hypoxia assessed by 18 F-misonidazole PET/CT (FMISO PET) and on multiparametric (mp) MRI parameters in HNSCC at an early (week 2) and late (week 5) time point during treatment and to analyse the relation between mpMRI and PET parameters with local control.