A Prostate-Specific Membrane Antigen PET-Based Approach for Improved Diagnosis of Prostate Cancer in Gleason Grade...

The preoperative Gleason grade group (GG) from transrectal ultrasound-guided prostate biopsy is crucial for treatment decisions but may underestimate the postoperative GG and miss clinically significant prostate cancer (csPCa), particularly in patients with biopsy GG1. In such patients, an SUV_max of at least 12 has 100% specificity for detecting csPCa. In patients with an SUV_max of less than 12, we aimed to develop a model to improve the diagnostic accuracy of csPCa. Methods: The study retrospectively included 56 prostate cancer patients with transrectal ultrasound-guided biopsy GG1 and an SUV_max of less than 12 from 2 tertiary hospitals. All [⁶⁸Ga]Ga-PSMA-HBED-CC PET scans were centrally reviewed in Xijing Hospital. A deep learning model was used to evaluate the overlap of SUV_max (size scale, 3 cm) and the level of Gleason pattern (size scale, 500 μm). A diagnostic model was developed using the PRIMARY score and SUV_max, and its discriminative performance and clinical utility were compared with other methods. The 5-fold cross-validation (repeated 1,000 times) was used for internal validation. Results: In patients with GG1 and an SUV_max of less than 12, significant prostate-specific membrane antigen (PSMA) histochemical score (H-score) H-score overlap occurred among benign gland, Gleason pattern 3, and Gleason pattern 4 lesions, causing SUV_max overlap between csPCa and non-csPCa. The model of 10 × PRIMARY score + 2 × SUV_max exhibited a higher area under the curve (AUC, 0.8359; 95% CI, 0.7233-0.9484) than that found using only the SUV_max (AUC, 0.7353; P = 0.048) or PRIMARY score (AUC, 0.7257; P = 0.009) for the cohort and a higher AUC (0.8364; 95% CI, 0.7114-0.9614) than that found using only the Prostate Imaging Reporting and Data System (PI-RADS) score of 5-4 versus 3-1 (AUC, 0.7036; P = 0.149) and the PI-RADS score of 5-3 versus 2-1 (AUC, 0.6373; P = 0.014) for a subgroup. The model reduced the misdiagnosis of the PI-RADS score of 5-4 versus 3-1 by 78.57% (11/14) and the PI-RADS score of 5-3 versus 2-1 by 77.78% (14/18). The internal validation showed that the mean 5-fold cross-validated AUC was 0.8357 (95% CI, 0.8357-0.8358). Conclusion: We preliminarily suggest that the model of 10 × PRIMARY score + 2 × SUV_max may enhance the diagnostic accuracy of csPCa in patients with biopsy GG1 and an SUV_max of less than 12 by maximizing PSMA information use, reducing the misdiagnosis of the PI-RADS score, and thereby aiding in making appropriate treatment decisions.