Weakly Supervised Learning-Driven Digital Pathology AI: A Systematic Review from Algorithmic Innovation to Clinical Translation

Abstract

Digital pathology is undergoing a paradigm shift from conventional microscopic diagnosis to computational pathology, with weakly supervised learning (WSL) serving as a key technology for reducing annotation costs. However, a significant gap remains between algorithmic innovation and clinical deployment. Addressing the core question of how WSL can advance digital pathology AI toward clinical use, this review systematically traces the technological evolution of WSL paradigms, clarifies the relationship between pathology foundation models and WSL, and examines the clinical applications and translational bottlenecks of weakly supervised pathology AI. The review indicates that while WSL substantially lowers the annotation cost barrier, the ultimate clinical value still depends on generalization capability, trustworthiness, and real-world validation.