Enhanced textural indices distinguish tissue types in FDG PET

Objectives PET textural indices (TI) measuring intratumoral heterogeneity show promising results for tumor characterization. Yet, recent studies demonstrated that TI were highly correlated with metabolic volume (MV) in small tumors. We propose a new resampling method for calculating TI in tumors of any volume without being biased by MV and demonstrate the ability of the resulting TI to differentiate between tissue types in FDG PET/CT.

Methods Forty-eight patients with non-small cell lung cancer underwent FDG PET/CT. Each primary lesion was segmented using an adaptive threshold and the resulting volume of interest (VOI) was duplicated in the liver of the patient. For each tumor and liver VOI, we resampled voxel intensities between 0 and 20 Standardized Uptake Value (SUV) units with 64 discrete values unlike the usual resampling (UR) that resamples between the minimum and maximum lesion SUV. We then computed 7 TI. ROC analyses were used to test the ability of TI to distinguish between tumor and liver tissue and between adenocarcinoma and squamous cell carcinoma.

Results Our new resampling (NR) cancelled the correlation between TI and MV. The Area Under the ROC Curve (AUC) for distinguishing between tissue types and between cancer subtypes was significantly higher with NR compared to UR for all 7 TI. Four NR TI (no UR TI) led to AUC statistically higher than that of SUVmax to distinguish between tumor and liver tissue (AUC ≥ 0.978 for NR TI vs AUC=0.930 for SUVmax). The cancer subtypes were best identified using Homogeneity and Low Gray-level Emphasis (AUC ≥ 0.756 for NR TI vs AUC=0.747 for SUVmax, p<0.05). Tumor tissues were more heterogeneous than liver tissue while squamous cell carcinomas were more heterogeneous than adenocarcinomas.

Conclusions NR method makes it possible to calculate TI that are independent from MV whatever the tumor volume. The NR TI are tissue-specific and might enrich the tumor profiling in radiomic analyses.