Back to all papers

Early prediction of response to ADT plus ARPIs in mHSPC using interpretable machine learning on PSMA PET/CT: a real-world study.

November 24, 2025pubmed logopapers

Authors

Wang Y,Yang Y,Qi L,Chen M,Hu S,Gao X,Tang Y,Cai Y

Affiliations (5)

  • Department of Urology, Disorders of Prostate Cancer Multidisciplinary Team, National Clinical Research Center for Geriatric Disorders, Xiangya Hospital, Central South University, Changsha, China.
  • Department of PET Center, National Clinical Research Center for Geriatric Disorders, Xiangya Hospital, Central South University, Changsha, China.
  • Department of Pathology, Disorders of Prostate Cancer Multidisciplinary Team, National Clinical Research Center for Geriatric Disorders, Xiangya Hospital, Central South University, Changsha, China. [email protected].
  • Department of PET Center, National Clinical Research Center for Geriatric Disorders, Xiangya Hospital, Central South University, Changsha, China. [email protected].
  • Department of Urology, Disorders of Prostate Cancer Multidisciplinary Team, National Clinical Research Center for Geriatric Disorders, Xiangya Hospital, Central South University, Changsha, China. [email protected].

Abstract

Reliable tools for early prediction of treatment response to androgen deprivation therapy (ADT) plus novel androgen receptor pathway inhibitors (ARPIs) in metastatic hormone-sensitive prostate cancer (mHSPC) remain lacking. This study aimed to develop and validate an interpretable machine learning model integrating [⁶⁸Ga]Ga-PSMA PET/CT-derived imaging features to predict response before therapy initiation. In this real-world study, 212 de novo mHSPC patients undergoing [⁶⁸Ga]Ga-PSMA PET/CT were included. Eighteen imaging and eight clinical features were extracted. Three ensemble-based recursive algorithms were applied for feature selection, and six machine learning models were developed. SHAP analysis was used for interpretability. Model outputs were combined with CHAARTED-defined tumor burden for refined risk stratification. Ten key features were identified, including metastatic total lesion PSMA uptake (m_TL_PSMA), whole-body peak standardized uptake value (w_SUVpeak), and primary lesion mean standardized uptake value (p_SUVmean). CatBoost achieved the best performance (training AUC = 0.908; test AUC = 0.904; Kappa = 0.717). In a temporally independent validation cohort, the model yielded an AUC of 0.875 (95% CI: 0.765-0.961), sensitivity of 0.72, and specificity of 0.89. Combined with CHAARTED classification, the model stratified patients into prognostic groups: responders with low-volume disease had the best metastatic castration-resistant prostate cancer (mCRPC)-free survival, while non-responders with high-volume disease had the worst outcomes. The explainable CatBoost-based model integrating quantitative PSMA PET/CT features demonstrated promising accuracy in predicting response to ADT plus ARPIs in mHSPC. When combined with the CHAARTED classification, it facilitated risk stratification. Prospective multicenter validation is required.

Topics

Journal Article

Ready to Sharpen Your Edge?

Join hundreds of your peers who rely on RadAI Slice. Get the essential weekly briefing that empowers you to navigate the future of radiology.

We respect your privacy. Unsubscribe at any time.