Abstract
290
Objectives: Introduction: Significant variations of SUVs and textural indices (TI) as a function of age in non-pathological breast tissue (NBT) have been reported and it has been suggested that this should be accounted for in future radiomic models assessing BC patients (Boughdad et al, JNM 2017;58;S1:470). It is also known that radiomic features vary according to the acquisition and reconstruction protocols. Our goal was to determine whether we could still detect the variations as a function of age when pooling data from two different centers and correcting the center-effect using a recently described harmonization
Methods: Methods: Patients who underwent 18F-FDG PET/CT at initial staging for various diseases excluding prior history of breast, ovarian or endometrial cancers were recruited retrospectively from two institutions (I1 and I2). The population of patients was classified into 3 age groups: childbearing age <45 yo (PRE); peri-menopausal/early menopausal women >45 and <55 yo (PERI) and post-menopausal women >55 and <85 yo (POST). In I1, patients were scanned on a Discovery 690 PET/CT scanner in a 3D mode using an ordered-subset expectation maximization (OSEM) iterative reconstruction algorithm (2 iterations, 24 subsets and voxel size: 2.7x2.7x3.3mm3). In I2, patients were scanned on a Biograph20_mCT PET/CT scanner in a 3D mode with Point Spread Function (PSF) modelling and Time Of Flight reconstruction (TOF, 2 iterations, 21 subsets and voxel size: 2x2x2mm3). For each patient in I1 and I2, a spherical Breast Volume Of Interest (B-VOI) >12mL was drawn in one breast and mirrored to the contralateral breast. Images were analyzed using LIFEx software (www.lifexsoft.org) to calculate SUV (mean, max, peak), Histogram-Based indices (HBI) and 6 TI (Homogeneity, Entropy, LRE, SRE, LGZE and HGZE) after resampling voxel intensities using 64 discrete values between 0 and 20 SUV units (bin width: 0.3). Anova tests were used to determine if SUV, HBI and TI values varied between age groups for I1 and I2 together and separately. GT2 Hochberg tests were used on post-hoc analysis. We also used t-tests to compare features values between I1 and I2 for each age groups (PRE, PERI and POST) before and after ComBat harmonization (Orlhac et al JNM 2018).
Results: Results: In I1, 327 patients with 654 B-VOI-I1 were included consisting of 71 pts PRE, 63 pts PERI and 193 pts POST. In I2, we had 96 patients with 192 B-VOI-I2 consisting of 15 pts PRE, 26 pts PERI and 55 pts POST. In both institutions separately, there were significant differences between age groups on Anova tests for all SUVs and most TI (except LGZE in I2) and on post-hoc analysis those differences were more pronounced between PRE vs POST and PERI vs POST groups. In the POST group, SUVmax, all HBI, SRE, LRE, LGZE and HGZE were significantly different between I1 and I2, demonstrating a center effect (p <0.05 ; t-test). This effect was also present in the PRE and PERI age groups for all HBI, in the PRE group for SRE and LRE and in the PERI group for SUVmax, LRE and HGZE (p <0.05 ; t-test). After ComBat harmonization, the significant differences in index between I1 and I2 for each age group were all removed, demonstrating the efficiency of ComBat. When pooling the I1 and I2 data, the age effect was present for all SUV, 2 HBI and 6 TI even without Combat (Anova, p<0.05). After ComBat, 11/13 p-values of Anova decreased demonstrating that real physiological effect (here, the age) was maintained and even enhanced after multi-center harmonization.
Conclusions: Conclusion: This study in 2 different institutions confirmed the existence of significant differences in SUV, HBI and TI in NBT as a function of age whatever the PET/CT system. ComBat harmonization allowed us to equalize PET biomarkers between PSF and non-PSF reconstructions algorithms without removing the influence of age. Multi-center radiomic studies of breast cancer can be performed using ComBat harmonization and the age covariate should be accounted for in the resulting radiomic models.