Abstract
453
Objectives Intratumor heterogeneity characterized by texture indices (TI) might be useful to predict treatment response in some types of cancer. Yet, it is unclear whether TI have added value compared to Standardized Uptake Values (SUV) or metabolic volume (MV) for histological classification. We studied whether TI improved tumor characterization with respect to conventional indices in NSCLC patients.
Methods Forty-eight patients with primary lesions of NSCLC underwent pre-surgery FDG PET. Histological analysis was performed on the tumor specimen after surgery to determine the TNM stage, subtype of cancer (adenocarcinoma vs epidermoid), presence of necrosis and differentiation. 6 TI were calculated from PET images in addition to SUV and MV. For each index, univariate analysis was performed with respect to the 5 histological criteria. The resulting statistically significant TI (p<5%) were combined with SUVmax or MV for multivariate analysis. Delong tests based on areas under the ROC curves (AUC) were used to determine if combined indices improved the histologic classification compared to conventional index alone.
Results In univariate analysis, 5 TI could predict all histological classifications (p<0.05) except the subtype of cancer. Only High Gray-level Zone Emphasis did not determine any histological classification. Among conventional index, MV best identified T and N stages (AUC=0.84 for T and 0.78 for N) and SUVmax best identified the subtype of cancer (AUC=0.75), the presence of necrosis (AUC=0.92) and differentiation (AUC=0.79). In multivariate analysis, no combination of indices was better than MV alone to characterize T and N stage or than SUVmax to identify necrosis and differentiation.
Conclusions In our NSCLC patients, tumor TI differ as a function of the histological classification. Yet, in multivariate analysis, no TI significantly improved the tumor characterization based on SUV or MV only.