Applications of Artificial Intelligence for Pathological Diagnosis of Bladder Cancer

The introduction and advancements of digital and computational pathology have made integration of Artificial Intelligence (AI) diagnostic algorithms into the clinical laboratory a reality. In the case of bladder cancer, a malignancy which places high burden both on patients and providers, AI has the potential to increase diagnostic sensitivity and improve diagnostic predictive performance. This review aims to synthesize current research and knowledge of the applications of AI in bladder cancer diagnostics. Overall, deep learning and convolutional neural network algorithms have successfully been applied to the diagnosis, grading, and differentiation of both non-muscle invasive and muscle-invasive bladder cancers. Additionally, AI shows great promise to improve treatment prediction and promotion of personalized medicine, including through integration of molecular signatures of bladder cancer. AI has also successfully been applied to cytopathology diagnostics, enabling quantitative and interactive identification of atypical cells. In order to promote clinical adaptation of these AI algorithms for the diagnosis of bladder cancer, key limitations must be addressed. This includes increasing interpretability of AI systems to address the black box effect, generation and collection of adequate and variable data sets, and proper consideration of realistic and safe integration into clinical practice. While AI has also been applied to areas of bladder cancer diagnosis such as cystoscopy and medical imaging, this is outside the scope of pathological diagnostics and is not discussed in this review.