Comparison of histogram-based and textural feature performance on FDG PET/CT and DCE-MRI to distinguish histological and immuno-histo-chemical parameters in breast cancer.

1365 Introduction: The use of texture analysis to assess relationship between tumor imaging phenotype and clinical, biological, histological and immuno-histo-chemical (IHC) parameters in breast cancer (BC) has shown promising results. To date, there are conflicting results on the usefulness of texture analysis on FDG PET/CT in BC (Soussan et al Plos One 2014, Lemarignier et al EJNMMI 2017). In Dynamic Contrast-Enhanced-MRI (DCE-MRI), the relevance of textural features has not been fully explored. In this work, we calculated textural features measured in a cohort of BC patients whom underwent FDG PET/CT and DCE-MRI at initial staging and investigated how they related to the histological and IHC tumor features. Methods: BC patients underwent at initial staging DCE-MRI with a dedicated bilateral breast coil (1.5 T) and PET/CT for which they received an IV injection of 3-3.5 MBq/kg of 18-FDG when capillary blood glucose was below 11 mMol/L and after 6 hours of fasting. In PET, texture indices (TI) were calculated after resampling voxel intensities using 64 discrete values (DV) between 0 and 20 SUV units (bin width: 0.3) after automatic segmentation of the tumor (40% of SUVmax). In DCE-MRI, 200 DV were used for resampling between 0 and 20000 units (bin width: 100) in the whole tumor manually delineated on DCE sequence obtained 2 minutes after Gadolinium IV injection. Six robust TI (Homogeneity, Entropy, SRE, LRE, LGZE and HGZE) and 4 histogram based-indices (HBI: skewnessH, kurtosisH, entropyH and EnergyH) for both PET and MRI were calculated in the PET and DCE-MRI tumor volumes. In PET, SUVs and Total Lesion Glycolysis (TLG) were also calculated. These feature values were analyzed as a function of the histological type: invasive ductal or lobular carcinoma (IDC or ILC), BC subtype (Luminal A, Luminal B HER 2-, Luminal B HER 2+, Basal-like and HER 2+) and tumor grade (III vs I+II) using multivariate and ROC analyses with a p value<0.05 for statistical significance. We also used Spearman correlation tests to assess the relationship between Ki-67 expression and tumor imaging features on PET and on DCE-MRI. Results: A population of 102 BC patients with a total of 110 tumors (8 pts had bifocal tumors) was recruited. In PET, on multivariate analyses there were significant differences between the histological types for 4 TI (homogeneity, entropy, LRE and HGZE) and 5 TI (homogeneity, entropy, LRE, LGZE and HGZE) were significantly different between the 5 BC subtypes. No significant differences were found in DCE-MRI TI between IDC and ILC or between BC subtypes. There were significant differences between tumor grades for 3 TI (entropy, SRE and HGZE) on F18-FDG PET/CT but only 1 TI (HGZE) on DCE-MRI (Table 1). HBI showed moderate performance to discriminate between histological types, BC subtypes or tumor grade on FDG PET/CT and very poor performance on DCE-MRI (Table 1). Using ROC analyses, SUVs and the 6 PET TI yielded AUC > 0.70 to distinguish between histological types with SUVmean having the highest AUC (0.77). PET LGZE yielded the highest AUC (0.80) to identify the tumor grade. In DCE-MRI, no image features could distinguish histological type or tumor grade with an AUC > 0.7. SUVs showed a higher correlation (0.53