Evaluation of 68ga-PSMA PET/CT with volumetric and texture analysis parameters for staging of prostate cancer patients

Aim: The aim of this study was to investigate the relationship between volumetric and radiomic data obtained from staging 68Ga-PSMA PET/CT images with PSA, risk groups, Gleason Grade (GG) groups and presence of metastasis. Method: Between January 2015 and September 2018, 88 patients who underwent 68Ga- PSMA PET-CT for staging of prostate adenocarcinoma were retrospectively analyzed. Images were evaluated in LIFEx program. PSMA involvement above the background activity in prostate gland, lymph node and other distant metastases were plotted with 40% threshold; radiomics & volumetric data were obtained. Results: In all patients, there was a moderate correlation between PSA and PSMA-TVtotal (p<0.001, r=0.580) and a high correlation between TL-PSMAtotal (p<0.001, r=0.636). PSMA-TVprostate and TL-PSMAprostate values were different in local and locally advanced/metastatic patients (p = 0.020 and 0.006, respectively). PSMA-TVprostate and TL- PSMAprostate values were significantly different in low-moderate and high-risk patients (p=0.003 and <0.001, respectively), and in patients with and without metastasis (p=0.008 and <0.001, respectively). PSMA-TVprostat-total, TL-PSMAprostat-total values were significantly different in patients with GG ≤3 and >3 (p=0.030, <0.001, 0.002 and <0.001, respectively). There was a low correlation between TL-PSMAprostate and GG groups (p=0.001, r=0.358), but there was no significant correlation between PSMA-TVprostate values and GG groups (p=0.137). There was low correlation between PSMA-TVtotal and moderate correlation between TL-PSMAtotal and GG groups in all patients (r =0.305 and 0.470, respectively). The texture analysis feature with the highest discriminating power to distinguish local and local advanced/metastatic groups, was GLZLM_HGZE (EAA=0.678). And also, the texture analysis parameter with the highest power to distinguish between medium and high risk groups according to D'Amico risk classification, was found to be GLZLM_HGZE (EAA=0.771). SUVpeak was the best discriminative SUV parameter for patients with GG 3≥ and> 3, and for high and low-moderate risk patients in the entire patient group and in the group without metastasis. AUC values were calculated with presence of metastasis and radiomic data, using naive bayes algorithm for predicting GG (>3 and 3≥) groups (model: 0.731, test: 0.753) and random forest algorithm for predicting GG (≥3 ve 3>) groups (model: 0.788, test: 0.727). Results: Without the use of invasive methods, the heterogeneity and aggressiveness of the primary tumor and the prediction of GG groups may be possible on PSMA PET/CT images obtained for diagnostic purposes; this holds promise for individualized disease management.