Application of artificial intelligence in kidney neoplasms: usability of pathological data in enhancing classification, grading and prognostic and predictive models

Abstract

Renal cell carcinoma (RCC) is among the most common human malignancies, gold standard in diagnosis is still histology but poses challenges in classification, grading, reproducibility or identification of predictive markers. The increasing use and availability of artificial intelligence (AI) like machine learning and deep learning methods, rose hope of improving those issues. The literature is expanding rapidly and in such experimental setting promising results were shown in distinguishing RCC subtypes and grades and leveraging digital pathology data in AI-integrated multimodal approaches combining histopathologic, genetic, and clinical data enhancing prognostic and predictive models. However, significant limitations hinder clinical implementation, like missing of prospective evaluation, underrepresentation of rare subtypes and evolving classification systems. Also the "black box" nature of some AI models and resource intensiveness raise concerns about transparency and feasibility.