Abstract
247
Objectives: PET/MRI hybrid imaging simultaneously provides morphologic and multiple functional data, which thus allows for voxel-by-voxel comparison of image data and for diversified characterization of neoplastic lesions. The aim of this study was to assess the correlation between standardized uptake value (SUV) of 18fluoreodeoxyglucose (18FDG) and apparent diffusion coefficient (ADC) of tumors of soft tissue and bone in a voxel-wise manner with a PET/MR hybrid scanner. MATERIALS AND
Methods: The consecutive 121 patients with tumor of soft tissue and bone were included in this study, and examined with 18FDG-PET/MR hybrid imaging (Ingenuity TF PET/MR; Philips Healthcare, Cleveland, OH). 3D-T2-weighted imaging (WI) with fat suppression, diffusion WI (DWI) with spatially selective radiofrequency excitation pulses (zoomed DWI, b = 0 and 800) and 18FDG-PET (3.7 MBq/kg FDG, 3-dimensional Ordered Subset Expectation Maximization + Time of Flight, 23-mm reconstruction) were simultaneously acquired. After image co-registration and reslice of 3D-T2WI and PET image based on ADC map on a workstation (Intellispace Portal 7.0, Philips Healthcare), values of SUV and ADC of the tumor were extracted on a voxel-by-voxel basis along with x-y coordinates. Scatter plots of SUV vs ADC and SUV/ADC vs logADC were generated for each tumor. Cluster analysis using k-means clustering algorithm was also applied to linear regression analysis on SUV/ADC vs logADC scatterplots when a tumor consists of multiple components such as liposarcoma, schwannoma and tumors with large necrotic portions by using our in-house program written with MATLAB 2015b (MathWorks, Natick, MA, USA). Pearson correlation coefficient was compared between SUV vs ADC and SUV/ADC vs logADC. The slopes of regression line were also compared among malignant, intermediate and benign groups. Statistical analysis was performed with Wilcoxon test, paired t-test, ANOVA and Mann-Whitney test. Results: According to the WHO classification 2013, malignancy grades of the 121 tumors were histologically classified into high-grade sarcomas (n=74), intermediate-grade (n=12) and benign tumor (n=35). The relationship between SUV/ADC and logADC showed a significant inverse linear correlation (mean r=-0.60, 95%CI: -0.63~-0.57), while that between SUV and ADC showed no significant correlation (mean r=-0.05, 95%CI: -0.10~0.00). All the three malignancy grades showed the significant difference of correlation coefficient between SUV/ADC vs logADC and SUV vs ADC (p<0.001). The slope was much steeper for SUV/ADC vs log ADC than SUV vs ADC in all the three grades. Cluster analysis was applied for 24 patients with liposarcoma (n=2), myxofibrosarcoma (n=3),schwannoma (n=2), other sarcomas (n=17). In all the 24 patients, both the two clusters (compartments) showed higher inverse correlation coefficients than the whole tumor Conclusion: Voxel-by-voxel SUV and ADC analyses with PET/MR hybrid systems can clarify the inverse linear correlation between SUV and ADC when ADC-corrected SUV was used as objective variable, which could derive a new quantitative imaging biomarker to characterize soft-tissue tumors. Captions Figure 1. Scatter plots of SUV vs ADC (a) and SUV/ADC vs logADC (b) in a patient with schwannoma Figure 2. Cluster analysis: FDG-PET/MR scans of a patient with dedifferentiated liposarcoma. Upper panel images are 3DT2WI(a),ADCmap(b) and FDG-PET fusion image (c). Lower panel mages are simple scatter plot of SUV vs ADC (d), semilog scatter plot of SUV/ADC vs ADC (e), and the semilog graph with cluster analysis (f).