Machine learning-based radiomics for multiple primary prostate cancer biological characteristics prediction with...

Area of Science:

• Nuclear Medicine
• Radiology
• Oncology

Background:

• ¹⁸F-PSMA-1007 PET radiomics can reveal primary prostate cancer (PCa) biological traits.
• Optimal SUVmax thresholds for these radiomics features remain undetermined.
• This study evaluates varying SUVmax thresholds to identify the best predictors for PCa characteristics.

Purpose of the Study:

• To compare the predictive accuracy of ¹⁸F-PSMA-1007 PET radiomics features at different SUVmax thresholds.
• To determine the optimal SUVmax thresholds for predicting key PCa biological characteristics.

Main Methods:

• 173 PCa patients' preoperative ¹⁸F-PSMA-1007 PET data were analyzed.
• Radiomics features were extracted from lesion volumes of interest at 30%, 40%, 50%, and 60% SUVmax thresholds.
• Support vector machine models predicted Gleason score (GS), extracapsular extension (ECE), and vascular invasion (VI), with performance assessed via ROC curves.

Main Results:

• The 50% SUVmax threshold yielded the best prediction for GS (AUC 0.80 in testing).
• The 40% SUVmax threshold was optimal for predicting ECE (AUC 0.77 in testing).
• The 50% SUVmax threshold also performed best for predicting VI (AUC 0.74 in testing).

Conclusions:

• ¹⁸F-PSMA-1007 PET radiomics features using 40-50% SUVmax thresholds demonstrate strong predictive performance for PCa biological characteristics.
• These radiomics features may offer added value beyond PSA for predicting PCa aggressiveness.