Remodeling of the periprostatic adipose microenvironment in aggressive prostate cancer: Insights from a multi-institutional atlas-based geometric analysis.

389 Background: Periprostatic adipose tissue (PPAT) plays an increasingly recognized role in prostate cancer progression. Spatial and morphological features of the periprostatic fat envelope on biparametric prostate MRI (bpMRI) have been associated with tumor aggressiveness. However, prior studies have relied on manual measurements from regions of interest. This study employed voxel-wise analysis to characterize morphological differences in the PPAT between patients with and without clinically significant prostate cancer (csPCA). Methods: A total of 556 patients with bpMRI from four institutions were analyzed. Periprostatic fat (PPF) masks on T2-weighted images were manually generated for 245 patients by an expert radiologist (8 years of experience) and used to train an nnU-Net v2 model (validation DSC = 0.85); the remaining cases were automatically segmented. A template (A - ) was constructed from 100 T2w images with iterative affine and deformable registration. All subjects were rigidly registered to A - space with their PPF masks and signed distance functions were computed. Voxel-wise group differences were assessed using permutation testing with threshold-free cluster enhancement and family-wise error (FWE) correction. From significant regions, twelve shape descriptors were extracted and compared between groups using Mann–Whitney U tests. Results: The cohort included 415 patients with csPCA (Gleason score ≥7) and 141 with benign or indolent disease (Gleason Grade Group=1). Median PSA was 7.82 ng/mL (range: 0.5–107.8) and median age was 69 years (range: 48–93). Voxel-wise analysis identified 1,569 significant voxels (p < 0.05, FWE-corrected), predominantly in the rectoprostatic region between the prostatic capsule and Denonvilliers' fascia. Three shape descriptors showed significant differences between groups. Patients with csPCA exhibited lower surface normal polar angle mean (0.683 vs. 0.787 rad, p = 0.007) and standard deviation (0.387 vs. 0.429 rad, p = 0.007), and lower mean curvature (0.0075 vs. 0.0094, p = 0.043). Conclusions: Patients with csPCA exhibited geometric differences in PPAT compared to those with benign or indolent prostate disease. Voxel-wise analysis identified these alterations in the adipose tissue between the prostatic capsule and Denonvilliers' fascia. These findings are consistent with studies indicating that PPAT is not merely a passive anatomical structure, but an active component of the tumor microenvironment. Significant radiomic shape descriptors of periprostatic fat. Feature What It Measures Benign/Indolent csPCA p-value θ mean Average tilt/orientation of fat surface around prostate 0.787 ± 0.428 0.683 ± 0.386 0.007 θ std Variability in fat surface orientation 0.429 ± 0.170 0.387 ± 0.162 0.007 κ mean Average curvature of fat boundary 0.009 ± 0.007 0.007 ± 0.007 0.043