Predictive role of Bladder wall dosimetry in conventional prostate Radiotherapy: Integrating AI and statistical Modelling.

BACKGROUND AND PURPOSE: Genitourinary toxicity is a common complication of prostate radiotherapy. Conventional bladder dosimetry may not adequately reflect the anatomical determinants of toxicity. This study evaluates the predictive value of bladder wall (BW) dosimetry and examines the added utility of artificial intelligence (AI) models. MATERIALS AND METHODS: A retrospective cohort of 177 prostate cancer patients treated with conventional radiotherapy was analyzed. Clinical, dosimetric, and toxicity data were collected. Logistic regression and machine learning (support vector machines, gradient boosting, and k-nearest neighbors) were employed to assess associations between bladder and BW parameters and acute urinary symptoms. RESULTS: BW-specific metrics were significantly associated with Grade ≥ 1 dysuria (BW V51Gy, p = 0.018) and pollakiuria (BW V65Gy, p = 0.030). Logistic regression yielded AUCs of 0.60-0.63, while machine learning-particularly gradient boosting-improved predictive performance (AUC up to 0.90 for dysuria and 0.97 for pollakiuria). Dose constraints of BW V51Gy < 5 cc and BW V65Gy < 3 cc were identified. CONCLUSIONS: Bladder wall dosimetry enriches the predictive landscape of genitourinary toxicity by accounting for anatomically precise dose distributions. AI-based modeling enhances risk stratification and may support the adoption of personalized dose constraints in clinical practice.