[Artificial intelligence as decision support tool in urological oncology: current evidence and challenges]

The rising incidence of cancer, increasing life expectancy and complex personalized treatment concepts pose considerable challenges for the healthcare system. Artificial intelligence (AI)-in particular machine learning (ML), deep learning (DL), and large language models (LLMs)-offers promising potential for supporting therapeutic decisions in urological oncology.

The aim of this review is to present the current state of research on the application of AI in treatment decisions in urological oncology. For this purpose, a systematic review of publications in the PubMed database was carried out. Methodological approaches, performance indicators and challenges with regard to clinical implementation were analysed comparatively.

Depending on the entity and treatment category, ML models achieve F1-scores between 0.75 and 0.99. Large language models that access external knowledge sources using retrieval-augmented generation (RAG) demonstrate a high degree of guideline compliance, as evidence-based knowledge is specifically integrated into the treatment recommendations. The explainability of the models is mainly ensured by Shapley additive explanation (SHAP) analyses or transparent guideline referencing.

To date, the available evidence predominantly consists of proof-of-concept studies, with a particular emphasis on prostate, urothelial, and renal cell carcinoma. Clinical implementation remains limited, in particular due to a lack of prospective validation studies, unresolved data protection and technical challenges, and limited model transparency and explainability.