A Prostate-Specific Membrane Antigen PET-Based Approach for Improved Diagnosis of Prostate Cancer in Gleason Grade Group 1: A Multicenter Retrospective Study

Flowchart of study design.

Overlap of SUV_max and overlap of PSMA intensity in patients with TRUS-guided biopsy GG1 and SUV_max < 12. Distribution of [⁶⁸Ga]PSMA PET/CT SUV_max demonstrated higher SUV_max in csPCa than in non-csPCa. However, between values of 2.9 (dashed gray line) and 9.2 (solid gray line), there was SUV_max overlap between csPCa and non-csPCa. Subsequently, 56 patients’ index-lesion slides corresponding to SUV_max were selected and made into 2 consecutive tissue sections. After this, 2 consecutive tissue sections were subjected to hematoxylin-eosin and PSMA immunohistochemistry staining and digital scanning. Resulting hematoxylin-eosin and PSMA images were segmented into 500 × 500 μm tiles. Deep learning was used to identify and cluster hematoxylin-eosin tiles, whereas PSMA tiles were matched with their corresponding hematoxylin-eosin tiles on basis of sample name and spatial coordinates of 2 consecutive tissue sections. Finally, according to categories of hematoxylin-eosin tiles, PSMA tiles were correspondingly labeled benign gland, GP3, and GP4 and randomly selected (n = 1,000 for each category) to calculate PSMA H-score using ImageJ software (National Institutes of Health). **P < 0.01 by Mann–Whitney test (overlap of SUV_max) and Kruskal–Wallis test (overlap of PSMA intensity). ****P < 0.0001 by Mann–Whitney test (overlap of SUV_max) and Kruskal–Wallis test (overlap of PSMA intensity). HE = hematoxylin-eosin.

Representative case of prostatic evasive anterior tumor shown on [⁶⁸Ga]PSMA PET/CT in axial, coronal, and sagittal views (from left to right). Representative case was previously underestimated as indolent PCa on basis of TRUS-guided biopsy result (biopsy GG1 and 1/13 positive core). [⁶⁸Ga]PSMA PET/CT (PRIMARY score, 4; SUV_max, 10.1) identified case as csPCa (RP GG2).

Relationships among SUV_max, PRIMARY score, and csPCa probability. (A) Nomogram of model of 10 × PRIMARY score + 2 × SUV_max for individual csPCa prediction. (B) Contour plot of csPCa probability according to 10 × PRIMARY score and 2 × SUV_max.