Predicting PSA50 response to [Formula: see text]Lu-PSMA therapy using machine learning and automated total tumor...

Area of Science:

• Nuclear Medicine
• Oncology
• Medical Imaging

Background:

• 177Lu-PSMA therapy is a standard treatment for metastatic castration-resistant prostate cancer (mCRPC).
• Predicting individual patient response to 177Lu-PSMA therapy remains challenging despite known correlations between 68Ga-PSMA PET/CT imaging and outcomes.
• Accurate response prediction is crucial for optimizing treatment strategies in mCRPC.

Purpose of the Study:

• To develop and validate an automated method for computing total tumor volume (TTV) from 68Ga-PSMA PET/CT imaging.
• To create predictive models for assessing patient biological response (PSA50) to 177Lu-PSMA therapy.
• To identify key imaging and clinical predictors of treatment response in mCRPC patients.

Main Methods:

• Retrospective analysis of 139 mCRPC patients treated with 177Lu-PSMA therapy.
• Automated extraction of TTV from 68Ga-PSMA PET/CT scans.
• Development of machine learning models (Logistic Regression with L1, SVM) to predict PSA50 response using TTV and clinical data.

Main Results:

• The automated TTV calculation and predictive models achieved F1-scores of 0.68 and 0.67, comparable to existing nomograms.
• TTV-derived features and time since diagnosis were identified as significant predictors of PSA50 response.
• The developed workflow provides a reproducible approach for predicting treatment outcomes.

Conclusions:

• An automated TTV measurement from 68Ga-PSMA PET/CT can aid in predicting 177Lu-PSMA therapy response in mCRPC.
• Machine learning models integrating imaging and clinical data enhance response prediction accuracy.
• Further refinement is needed for lesion segmentation in physiological areas to improve model robustness.