Uptake time dependency of Texture Analysis on Dual Phase FDG PET for Soft Tissue Sarcomas

Objectives: Soft tissue sarcomas (STS) are relatively rare types of tumors but have variable levels of tumor differentiation. STS are known to have biological heterogeneity associated with cell proliferation, necrosis, vascularity and metabolic activities. FDG PET-CT has been an established method to evaluate STS, and STS often have heterogeneous uptake of FDG. Texture analysis is a group of computational methods that can quantify heterogeneity to characterize tumor. SUV is known to increase over time after FDG injection in malignant lesion. Metabolic tumor volume (MTV) and total lesion glycolysis (TLG) also change over time. Therefore, to compare two studies within or between subjects, the uptake time needs to be identical between studies, which is often difficult in clinical settings. This study investigates time dependency of textural features between early vs. delayed acquisition in comparison with conventional PET parameters.

Methods: In this retrospective study, we investigated 17 patients (59±19 years old, range 18 - 87 years old) with pathologically confirmed STS. All the patients underwent FDG PET before treatment. The patients fasted at least 6 hours before examination. Patients were injected with FDG (220±64 MBq) once, followed by dual phase scanning; at an early phase (61.5±2.6 min) and at a delayed phase (117.9±2.2 min) using a standalone PET scanner (SHIMADZU Eminence SET3000-G). Images were reconstructed with ordered subset expectation maximization algorism. The voxel size of all images were set to 4.0×4.0×2.0 mm. A total of 34 FDG-PET datasets (early and delayed images, from 17 patients) were processed as follows. Tumor boundary was defined using a fixed threshold of SUV 蠅 2.0. Histogram and 4 matrices were generated in a 3-dimentional manner from all the voxels within the tumor, calculating conventional PET parameters (SUVmax, SUVmean, MTV, and TLG) and 36 textural features using R version 3.3.2. Textural features and conventional PET parameters from early and delayed images were compared using intra-class correlation (ICC) and paired t-test.

Results: From early to delayed phases, 31 out of 36 texture features did not differ significantly (p=N.S., t-test), although 3 of 4 conventional PET parameters increased significantly (SUVmax and SUVmean; p<0.001, TLG; p<0.05). In Textural features, 3 features increased significantly, and 2 features decreased significantly. 34 out of 36 textural features showed high ICC (0.99>ICC>0.82) between early phase and delayed phase. 2 out of 36 textural features did not differ significantly although showing low ICC. Based on no significant differences and high ICC between early and delayed phases, 29 out of 36 textural features were considered least sensitive to uptake time.

Conclusion: In texture analysis of dual phase FDG PET on STS, 29 out of 36 texture features did not differ significantly between early and delayed phases. ICC of the features was high. These textural features were more robust and less sensitive to the uptake time than SUVmax and SUVmean. In conclusion, texture analysis of FDG uptake may be a new and reliable method when analyzing multi-center datasets with a fluctuating uptake time of FDG PET. Research Support: