Development, multi-institutional external validation, and algorithmic audit of an artificial intelligence-based...

Area of Science:

• Urology
• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate prediction of side-specific extraprostatic extension (ssEPE) is crucial for nerve-sparing surgery in localized prostate cancer.
• AI offers potential for personalized ssEPE predictions to guide surgical strategies and minimize side effects like impotence and incontinence.

Purpose of the Study:

• To develop, externally validate, and audit an AI tool, SEPERA (Side-specific Extra-Prostatic Extension Risk Assessment), for predicting ssEPE.
• To assess SEPERA's performance against existing nomograms and logistic regression models.

Main Methods:

• SEPERA was trained on 1022 cases and validated on 3914 cases across three international academic centers.
• Performance was evaluated using AUROC, AUPRC, calibration, and net benefit, with comparisons to contemporary nomograms and a logistic regression model.
• An algorithmic audit assessed model bias across various patient demographics and clinical factors.

Main Results:

• SEPERA demonstrated strong performance (pooled AUROC 0.77) and calibration across validation cohorts.
• It outperformed other models in predicting ssEPE, particularly in challenging cases, and showed higher net benefit.
• The algorithmic audit revealed no significant model bias across diverse patient groups; common errors were false positives in older, high-risk patients.

Conclusions:

• The AI-based SEPERA tool is accurate, safe, and generalizable for personalized nerve-sparing strategies in radical prostatectomy.
• SEPERA can enhance surgical decision-making, potentially improving outcomes and reducing treatment-related side effects.