Microenvironment at a Distance: Multi-Endocrine-Organ Radiomics to Identify Systemic Signatures in PSMA-Negative...

Background/Introduction: Prostate cancer (PCa) is the most commonly diagnosed malignancy among men and remains a major cause of cancer-related mortality worldwide. We aimed to evaluate whether radiomic features extracted from normal endocrine organs, combined with clinical variables, could predict clinical progression in patients with PSMA-negative prostate cancer. Materials and Methods: In this retrospective study, 101 men with biochemically recurrent prostate cancer and negative [¹⁸F]DCFPyL PET/CT scans were included. Radiomic features were extracted from the adrenal glands, thyroid, the hypothalamus-pituitary complex, and testes. Post-imaging variables were excluded to prevent temporal data leakage. Models were developed using a stratified train/test split framework with preprocessing and feature selection performed exclusively within the training subset prior to evaluation on the held-out test set. Performance was evaluated using AUC, accuracy, sensitivity, specificity, and Brier score, while bootstrap confidence intervals and DeLong analysis were used for statistical assessment. Results: Multimodal fusion models integrating CT radiomics, PET radiomics, and clinical variables demonstrated the strongest predictive performance. The highest-performing model combined TESTIS_CT and TESTIS_PET radiomics with clinical variables, achieving an AUC of 0.758 (95% CI: 0.653-0.849). Clinical-only models remained highly competitive, with the best configuration achieving an AUC of 0.727 (95% CI: 0.618-0.833). PET + clinical and CT + clinical models achieved AUC values of up to 0.733 and 0.729, respectively, while imaging-only models demonstrated substantially lower discrimination. Although endocrine organ radiomics numerically improved predictive performance and specificity, DeLong analysis demonstrated no statistically significant improvement beyond clinical variables alone. Discussion: These findings suggest that endocrine organ radiomics may provide complementary system-level imaging biomarkers reflecting tumor-host interactions in PSMA-negative prostate cancer. However, their incremental clinical value remains modest. Conclusions: Endocrine organ radiomics combined with clinical variables demonstrated promising predictive performance in PSMA-negative prostate cancer, particularly in multimodal fusion models. Nevertheless, the added value beyond clinical variables alone was not statistically significant and requires validation in larger independent cohorts.