What question did this study set out to answer?

The research aims to evaluate radiomic features from [18F] PSMA-3Q PET/CT to predict post-surgical ISUP grades and extraprostatic extension in prostate cancer management.

June 8, 2026Cancer Imaging0 citationsOpen Access

18F PSMA-3Q PET/CT radiomics at 45% SUVmax threshold: predicting post-surgical ISUP grade ≥ 4 and extraprostatic extension for noninvasive stratification in prostate cancer management

JCJianbo CuiChinese PLA General Hospital YLY LiuChinese PLA General Hospital GWGuanyun WangCapital Medical University

Key Points

The research aims to evaluate radiomic features from [18F] PSMA-3Q PET/CT to predict post-surgical ISUP grades and extraprostatic extension in prostate cancer management.
243 prostate cancer patients underwent [18F] PSMA-3Q PET/CT before radical prostatectomy, divided into training (n=190) and test (n=53) cohorts.
Radiomics features extracted with 40%, 45%, and 50% SUVmax thresholds; nine machine learning algorithms were utilized.
Model performance evaluated through area under the curve (AUC), sensitivity, specificity, positive and negative predictive values.
PC45 model achieved an AUC of 0.917 (95% CI: 0.828–0.985) for predicting psISUP grade ≥ 4, with 0.800 sensitivity and 0.921 specificity.
For EPE prediction, the AUC was 0.772 (95% CI: 0.633–0.883), sensitivity 0.700, and specificity 0.727.
The model demonstrated stable performance across different PSA density subgroups.

Abstract

To systematically evaluate radiomic features extracted from 18 F PSMA-3Q PET/CT using 40%, 45%, and 50% SUVmax thresholds for their ability to predict post-surgical International Society of Urological Pathology (psISUP) grading and extraprostatic extension (EPE) in prostate cancer and ultimately develop an optimal threshold-based predictive model. This retrospective study included 243 prostate cancer patients undergoing 18 F PSMA-3Q PET/CT before radical prostatectomy. Patients were chronologically divided into training ( n = 190) and test ( n = 53) cohorts, with stratification by prostate-specific antigen density (PSAD < 0.25 vs. ≥ 0.25 ng/mL). Radiomics features were extracted using 40%, 45%, and 50% SUVmax thresholds. Nine machine learning algorithms developed integrated PET/CT radiomics-clinical models (PC40, PC45, PC50) and PET radiomics-clinical models (P40, P45, P50). Model performance was evaluated using area under the curve (AUC), sensitivity, specificity, positive and negative predictive values. Among the models constructed with 40%, 45%, and 50% SUVmax thresholds, the PC45 model (PET/CT radiomics + clinical parameters) performed best. For predicting psISUP grade ≥ 4, its test set AUC reached 0.917 (95% CI: 0.828–0.985), with 0.800 sensitivity and 0.921 specificity. For EPE prediction, the test set AUC was 0.772(95% CI: 0.633–0.883), sensitivity 0.700 and specificity 0.727. It showed stable performance across PSAD subgroups and higher net clinical benefit. The 18 F PSMA-3Q PET/CT radiomics model with a 45% SUVmax threshold may facilitate relatively accurate noninvasive prediction of psISUP grade and EPE. These findings suggest its potential value as a promising preoperative tool for precision prostate cancer management, and provide preliminary evidence supporting its stable and balanced performance as a noninvasive decision‑making aid, pending further validation in larger cohorts.

Ask AI

Helpful

Bookmark

View Full Paper