Predicting Response to [177Lu]Lu-PSMA Therapy in mCRPC Using Machine Learning.

Area of Science:

• Oncology
• Nuclear Medicine
• Medical Imaging

Background:

• Metastatic castration-resistant prostate cancer (mCRPC) treatment with [177Lu]Lu-PSMA radioligand therapy (RLT) is emerging.
• Patient response and tolerance to RLT vary, necessitating identification of predictive factors.
• Distinguishing fully beneficial treatment patients (FBTP) from not fully beneficial treatment patients (NFBTP) is crucial for optimizing RLT outcomes.

Purpose of the Study:

• To identify predictive factors for distinguishing FBTP from NFBTP undergoing RLT for mCRPC.
• To evaluate the role of machine learning in predicting RLT treatment effectiveness.
• To enhance patient selection for personalized RLT treatment planning.

Main Methods:

• Analysis of clinical, imaging, and biological data from 25 mCRPC patients (11 FBTP, 14 NFBTP).
• Application of exploratory data analysis and feature engineering for machine learning model development.
• Integration of Choline PET imaging and blood parameters (neutrophils, leukocytes, alkaline phosphatase) to improve predictive accuracy.

Main Results:

• Significant differences in renal Choline uptake intensity observed between FBTP and NFBTP.
• Discordance between FDG PET and PSMA imaging (FDG+/PSMA-) identified as a potential indicator for NFBTP.
• Machine learning model integrating imaging and biological data achieved 0.92 accuracy, 0.96 sensitivity, and 0.96 precision.

Conclusions:

• An integrated approach combining imaging (especially Choline PET) and biological data significantly enhances predictive accuracy for RLT outcomes in mCRPC.
• Machine learning effectively refines patient selection for RLT, optimizing treatment planning.
• Identifying predictive factors improves understanding of RLT efficacy and patient stratification.