Radiomic analysis of pre-treatment prostate-specific membrane antigen (PSMA) PET/CT for prognostic stratification in patients with oligometastatic castration-sensitive prostate cancer

Abstract

Objective

To explore the imaging biomarkers obtained from baseline multi-modality imaging PET/CT before metastasis-directed therapy (MDT), which could offer early response prediction before MDT treatment, optimizing patient management and improving outcomes.

Methods

The study analyzed a multi-institutional cohort of 118 patients with oligometastatic castration-sensitive prostate cancer (omCSPC), including 34 from Johns Hopkins Hospital (JHH) and 84 from Baskent University (BU), USA, all treated with stereotactic ablative radiation therapy SABR-MDT. Before MDT, all patients underwent PSMA PET and CT imaging. For radiomics analysis, the gross tumor volume (GTV) was defined as zone 1, with an additional 5 mm peritumoral expansion designated as zone 2. From these regions, 1308 radiomics features were extracted. Feature selection was performed using a mutual information function, identifying the five most informative radiomics features from prostate-specific membrane antigen (PSMA) PET and CT. These were combined with five key clinical parameters—age, Gleason score, total number of lesions, number of untreated lesions, ADT, and pre-MDT prostate-specific antigen (PSA)—as model inputs. Multiple machine-learning algorithms, including random forest, decision tree, support vector machine, and naïve Bayes, were applied to predict 2-year metastasis-free survival (MFS). Model performance was evaluated using both leave-one-out and cross-institution validation.

Results

In a leave-one-out test with 93 patients, random forest achieved 78% accuracy and an AUC of 0.80 in predicting 2-year MFS. In cross-institution validation with 61 BU and 32 JHH patients, random forest correctly predicted 2-year MFS for 69% and 71% of patients, with AUC values of 0.71 and 0.73, respectively. Kaplan Meier curve comparison shows statistically significant separation between “rapid progressors” and “non-rapid progressors” patients stratified by the model in both leave one out and cross-institution validation tests.

Conclusion

This study provides evidence that pre-treatment multi-modality imaging biomarkers derived from PSMA PET and CT can serve as valuable predictors of metastasis-free survival (MFS) in patients with omCSPC.