A Spatial Perspective on Extraprostatic Extension in Prostate Cancer: Multiscale Geometric Analysis of Multiparametric MRI.

RATIONALE AND OBJECTIVES: To develop an objective, reproducible approach for preoperative assessment of extraprostatic extension (EPE) by characterizing multiscale geometric phenotypes on multiparametric MRI (mpMRI) across lesion, gland, and tumor-capsule interface levels. MATERIALS AND METHODS: In this retrospective diagnostic accuracy study, 138 patients with prostate cancer undergoing radical prostatectomy within 180 days after mpMRI were included. Tumor and prostate segmentations yielded 2D and 3D descriptors of lesion shape, gland morphology, and tumor-capsule interaction. Reproducibility, assessed by intraclass correlation coefficients, guided selection of descriptors integrated via multivariable logistic regression. The cohort was divided into training and test sets, and performance evaluated by area under the receiver operating characteristic curve (AUC), calibration (Hosmer-Lemeshow test), and decision curve analysis. RESULTS: The cohort included 138 men (mean age, 70 ± 7 years), of whom 80 (57.97%) had pathologic EPE. EPE-positive tumors showed larger size, lower 3D sphericity, higher fractal dimension (P < 0.05), greater tumor-capsule interaction, and subtle prostate geometry changes (P < 0.05). A compact geometric model integrating lesion sphericity, prostate flatness, and convexity achieved AUCs of 0.780 (95% CI: 0.678-0.882) and 0.771 (95% CI: 0.645-0.897) in the training and test sets, comparable to expert EPE grading (P > 0.05). CONCLUSION: A multiscale geometric framework integrating 2D and 3D descriptors captures complementary spatial phenotypes associated with EPE on mpMRI. Geometry-based modeling achieved discrimination comparable to expert assessment, providing an objective and reproducible adjunct for preoperative evaluation, particularly where subspecialty expertise may be limited.