Machine Learning-based Dose Prediction in [¹⁷⁷Lu]Lu-PSMA-617 Therapy by Integrating Biomarkers and Radiomic Features...

Area of Science:

• Nuclear Medicine
• Radiopharmaceutical Therapy
• Medical Imaging

Background:

• Metastatic castration-resistant prostate cancer (mCRPC) treatment often involves radioligand therapy (RLT).
• Accurate dosimetry is crucial for optimizing RLT efficacy and minimizing toxicity.
• Predicting absorbed doses (ADs) before treatment can aid in patient selection and treatment planning.

Purpose of the Study:

• To develop machine learning (ML) models for predicting pretherapy absorbed doses (ADs) in kidneys and tumoral lesions.
• To utilize radiomic features (RFs) from [68Ga]Ga-PSMA-11 (Ga-PSMA) PET/CT scans and clinical biomarkers (CBs).
• To potentially improve patient selection and enable dosimetry-guided therapy for [177Lu]Lu-PSMA-617 (Lu-PSMA) RLT.

Main Methods:

• Twenty mCRPC patients underwent Ga-PSMA PET/CT scans prior to Lu-PSMA RLT.
• Posttherapy dosimetry was performed using scintigraphy and SPECT/CT imaging.
• Monte Carlo simulations derived ADs, with ML models trained on pretherapy RFs and CBs.

Main Results:

• The best ML model for kidney AD prediction (extra trees regressor) achieved an R² of 0.87 and RMSE of 0.11 Gy/GBq.
• The best ML model for lesion AD prediction (gradient-boosting regressor) achieved an R² of 0.77 and RMSE of 1.04 Gy/GBq.
• Combining clinical biomarkers with radiomic features yielded optimal prediction results.

Conclusions:

• Integrating pretherapy Ga-PSMA PET/CT radiomic features with clinical biomarkers shows promise for predicting absorbed doses in RLT.
• These findings suggest potential for personalized treatment planning and enhanced patient stratification.
• Larger sample sizes and independent cohort validation are recommended for further confirmation.