Advances in computational nephropathology

Abstract

Pathology relies on pathologists' qualitative assessment and semiquantitative measures to characterize the structural and molecular alterations of tissues. Novel analytical methods and recent advances in the computational field, particularly in artificial intelligence and deep learning in pathology, termed computational pathology, have led to widespread applications and advancement in research. Integrating computational approaches into the digital pathology workflow can facilitate the automated, high-throughput analysis of histopathologic images, thereby improving precision, reproducibility, and efficiency in pathology diagnostics. We provide a comprehensive overview of the advancements and applications of computational pathology, specifically in nephropathology. We discuss widely adopted methodological approaches, highlighting their respective strengths and limitations, including quantitative nephropathology (i.e., pathomics), deep learning-based image classification and regression, and nonimage applications (e.g., automated decision support systems for standardizing the reporting of current consensus classifications). Despite the promising potential of these approaches, several challenges remain for successful implementation in routine clinical practice. We highlight technological, regulatory, and ethical challenges, such as computational infrastructure, data privacy, and considerations of environmental sustainability. Looking toward the future, we envisage potential developments that could further transform the field. We are entering a new exciting era, where computational methods are reshaping and redefining kidney pathology, perhaps also renaming our field to "kidnAI" pathology.

Keywords: deep learning; digital pathology; kidney biopsy; pathomics.