# Preoperative Multiparametric MRI‐Based Tumour–Periprostatic Adipose Tissue Interface Characterisation for Extraprostatic Extension Prediction in Prostate Cancer

## TL;DR

This study shows that analyzing the interface between prostate tumors and surrounding fat tissue on MRI improves the prediction of cancer spread beyond the prostate.

## Contribution

Simplified tumor–PPAT interface features on MRI enhance EPE prediction without increasing imaging complexity.

## Key findings

- The combined model with interface features achieved an AUC of 0.823 for EPE prediction.
- The combined model showed higher net benefit than the baseline model across threshold probabilities.
- Interface features independently predicted EPE and improved preoperative risk stratification.

## Abstract

To evaluate the independent predictive value of tumour–periprostatic adipose tissue (PPAT) interface features on preoperative multiparametric magnetic resonance imaging (mpMRI) for extraprostatic extension (EPE) in prostate cancer and to compare discrimination and clinical net benefit with a baseline clinical model.

This single‐centre retrospective cohort included patients who underwent radical prostatectomy with mpMRI completed within 8 weeks. On a single axial slice at maximum tumour diameter, five simplified interface features were measured using standard PACS tools: contact length, contact angle, T2 signal intensity ratio, interface apparent diffusion coefficient (3‐mm annular zone) and capsular integrity score (0–2 scale). A baseline clinical model (prostate‐specific antigen [PSA], PSA density, PI‐RADS and biopsy Gleason score) and a combined model (baseline variables plus LASSO‐selected interface features) were constructed. Bootstrap internal validation (1000 iterations) with bias correction was performed. Discrimination was assessed using the area under the curve (AUC), and calibration curves and decision curve analysis evaluated accuracy and net clinical benefit.

A total of 240 patients were included, with an EPE prevalence of 34.2% (82/240). The combined model achieved a bias‐corrected AUC of 0.823 (95% confidence interval [CI]: 0.768–0.878), suggesting improvement over the baseline model's AUC of 0.744 (95% CI: 0.680–0.808). Decision curve analysis revealed a higher net benefit for the combined model across clinically relevant threshold probabilities (10%–50%).

Simplified tumour–PPAT interface features independently predict EPE without increasing imaging complexity, improving discrimination and clinical value for preoperative risk stratification.