18F-Fluoroethylcholine PET/CT Radiomic Analysis for Newly Diagnosed Prostate Cancer Patients: A Monocentric Study

Aim: The aim of this study is to assess whether there are some correlations between radiomics and baseline clinical-biological data of prostate cancer (PC) patients using Fluorine-18 Fluoroethylcholine (F-18-FECh) PET/CT. Methods: Digital rectal examination results (DRE), Prostate-Specific Antigen (PSA) serum levels, and bioptical-Gleason Score (GS) were retrospectively collected in newly diagnosed PC patients and considered as outcomes of PC. Thereafter, Volumes of interest (VOI) encompassing the prostate of each patient were drawn to extract conventional and radiomic PET features. Radiomic bivariate models were set up using the most statistically relevant features and then trained/tested with a cross-fold validation test. The best bivariate models were expressed by mean and standard deviation to the normal area under the receiver operating characteristic curves (mAUC, sdAUC). Results: Semiquantitative and radiomic analyses were performed on 67 consecutive patients. tSUVmean and tSkewness were significant DRE predictors at univariate analysis (OR 1.52 [1.01; 2.29], p = 0.047; OR 0.21 [0.07; 0.65], p = 0.007, respectively); moreover, tKurtosis was an independent DRE predictor at multivariate analysis (OR 0.64 [0.42; 0.96], p = 0.03) Among the most relevant bivariate models, szm_2.5D.z.entr + cm.clust.tend was a predictor of PSA levels (mAUC 0.83 +/- 0.19); stat.kurt + stat.entropy predicted DRE (mAUC 0.79 +/- 0.10); cm.info.corr.1 + szm_2.5D.szhge predicted GS (mAUC 0.78 +/- 0.16). Conclusions: tSUVmean, tSkewness, and tKurtosis were predictors of DRE results only, while none of the PET parameters predicted PSA or GS significantly; F-18-FECh PET/CT radiomic models should be tested in larger cohort studies of newly diagnosed PC patients.