Abstract

Evangelin Gabriel Rajesh Kumar, Monisha Krishnakumar, *Lashika L. K.

ABSTRACT

Background: Non-muscle invasive bladder cancer (NMIBC) constitutes the majority of bladder cancer cases and is primarily treated with intravesical Bacillus Calmette-Guérin (BCG) immunotherapy. However, up to 40% of patients do not respond to BCG, highlighting the need for reliable predictive tools to guide personalized treatment strategies. Artificial intelligence (AI) has emerged as a promising approach to address this challenge by integrating complex clinical and biological data. Methods: A comprehensive literature search was conducted across multiple databases up to November 2025. Studies evaluating AI techniques—including deep learning, machine learning, radiomics, and multimodal models—for predicting BCG response were included. Data on model performance, input modalities, and clinical applicability were analyzed using narrative synthesis. Results: AI-based models demonstrated superior predictive performance compared to traditional clinical risk stratification tools. Deep learning models showed high accuracy, particularly with histopathological and genomic data, though interpretability remains limited. Classical machine learning approaches offered improved transparency with comparable performance. Radiomics enabled non-invasive prediction of tumor characteristics and immune microenvironment features. Multimodal models integrating imaging, pathology, molecular, and clinical data consistently achieved the highest predictive accuracy. Certain AI models also demonstrated potential in guiding treatment selection and identifying patients unlikely to benefit from BCG therapy. Conclusion: AI-based predictive models hold significant potential for improving treatment decision-making and enabling personalized therapy in NMIBC. However, challenges such as limited prospective validation, data heterogeneity, and lack of standardization must be addressed before routine clinical implementation.