Relationship between Prostate-Specific Antigen Kinetics and Radiomic Features of F-18 Fluorocholine PET in Patients with Prostate Cancer

3004 Objectives: Prostate-specific antigen (PSA) levels and kinetics are widely used in the management of prostate cancer. F-18 fluorocholine (FCH) PET imaging can be helpful in the evaluation of patients with prostate cancer. The aim of this work was to examine the relationship of PSA levels and kinetics with FCH PET radiomic features in prostate cancer patients. Methods: Eighteen patients with prostate adenocarcinoma were included in this retrospective study. The radiomic features of FCH PET were correlated with PSA level in 18 patients, with PSA doubling time (PSAdt) in 10 patients, with PSA velocity (PSAvel) in 10 patients, and with Gleason score in 18 patients. FCH-positive lesions on PET images were contoured and segmented by applying the region growing algorithm implemented in Osirix software (Pixmeo, Geneva, Switzerland) with a 42% threshold of maximum standard uptake value (SUV). A total of 72 heterogeneity indices containing 13 first-order intensity features, and 59 second- or higher-order textural features were computed by using the open-source software package Chang-Gung Image Texture Analysis Toolbox. A heat map was generated to display the correlations of PET radiomics with PSA level, PSAdt, PSAvel, and Gleason score. Correlation coefficients were calculated using Spearman9s rank order test. All statistical tests were two-sided with a significance level of 0.05. Results: Forty heterogeneity features were significantly associated with PSA level. Among these features, run percentage (ρ = -0.736, p < 0.01), zone percentage (ρ = -0.717, p < 0.01), high-intensity zone emphasis (ρ = -0.761, p < 0.01), high-intensity short-zone emphasis (ρ = -0.742, p < 0.01), high-intensity large-zone emphasis (ρ = -0.707, p < 0.01), SUV kurtosis (ρ = 0.746, p < 0.01), and total lesion activity (TLA) (ρ = 0.723, p < 0.01) showed high correlation coefficients. Three heterogeneity indices were significantly related to PSAdt, and run percentage (ρ = 0.778, p < 0.01) demonstrated a high correlation. Among the 11 features with significant correlations with PSAvel, run percentage (ρ = -0.888, p < 0.01), zone percentage (ρ = -0.745, p = 0.013), high-intensity zone emphasis (ρ = -0.745, p = 0.013), high-intensity short-zone emphasis (ρ = -0.709, p = 0.022), high-intensity large-zone emphasis (ρ = -0.758, p = 0.011), SUV kurtosis (ρ = 0.830, p < 0.01), TLA (ρ = 0.782, p < 0.01), and peak SUV lean body mass (SUL) (ρ = 0.758, p = 0.011) exhibited high correlations. One index, peak SUL (ρ = 0.511, p = 0.03), was significantly and moderately related to Gleason score. The correlations of radiomic features and PSA-related parameters are visualized in the supplementary figure. Conclusions: This preliminary study revealed that the radiomic features of FCH PET images have a good relationship with PSA-related parameters. Further works include examining the usefulness of these signatures in a large cohort population.