Harmonization of SUV and Voxel-Size Reduces the Scanner-dependent Biases in Intratumor Heterogeneity Indices from F-18 FDG PET Images Obtained by Various Scanners

1778 Objectives We aim to investigate the effect of harmonization of SUV and voxel-size to the biases of intratumor heterogeneity indices from F-18 FDG PET images obtained by various scanners with different spatial resolution and voxel-size. Methods We retrospectively analyzed preoperative F-18 FDG PET scans from 97 patients for the diagnosis of thymic epithelial tumors (TETs). TETs were divided into three subgroups according to increasing degree of malignancy; low-risk thymoma (LRT), high-risk thymoma (HRT), and thymic carcinoma (TC). Forty seven PET images were acquired on two high-resolution PET scanners with point spread function (PSF)-based reconstruction (Discovery6 90/Biograph Truepoint 40), whereas the other 50 PET images were acquired on a low resolution (non-PSF) PET scanner (Biograph Sensation 16). We harmonized SUV and voxel-size of PET images from higher resolution scanner to that of the lowest, using ACR phantom-based recovery coefficient matching method (Oh et al., 2014). After obtaining SUV and tumor heterogeneity indices (4 local-level and 16 region-level indices) of TETs before and after harmonization, we regressed them by TET subgroups and PET scanner groups (high/low resolution). To investigate if the harmonization reduces scanner-induced differences of SUV and heterogeneity indices, we compared significances of the scanner-group variables between models using harmonized and non-harmonized indices. Results Twenty (95%) out of 21 indices (SUVmax and 20 intratumor heterogeneity indices) before harmonization showed significant scanner dependency (P 0.05) in 10 out of21 indices. Conclusions By the SUV and voxel-size harmonization, the PET scanner dependency of SUV and intratumor heterogeneity indices could be reduced. The harmonization step should be preferred when using data from various PET scanners with different spatial resolution and voxel-size. Research Support This study was supported by a grant of the Korean Health Technology R&D Project, Ministry of Health & Welfare, Republic of Korea. (HI07C0001).