Rising caseloads, diagnostic complexity, and workforce shortages are challenging traditional microscope-based pathology practice. Digital pathology (DP) and artificial intelligence (AI) have emerged as transformative solutions.
DP infrastructure includes whole-slide scanners, image management systems, tiered storage architectures, and high-performance networks, integrated with laboratory information systems. The 2017 FDA approval of the first WSI system for primary diagnosis was a clinical milestone. Most institutions implement DP in phases, typically proceeding through pilot, hybrid, and full-digital stages, requiring diagnostic concordance validation and ongoing quality assurance.
AI in pathology has evolved from traditional machine learning to deep learning, and now to multimodal foundation models. Key clinical applications include: (1) classification and diagnosis—Gleason grading and mitosis detection achieving performance comparable to experts; (2) screening and detection—lymph node metastasis detection (AUC up to 0.994) and microcalcification identification; (3) biomarker quantification—automated assessment of ER/PR, HER2, Ki-67, and PD-L1; (4) tumor grading; (5) tumor microenvironment evaluation; and (6) prognosis and treatment response prediction—predicting recurrence risk and molecular subtypes from H&E slides.
Emerging frontiers include multimodal foundation models (integrating images, text, and genomics for cross-modal reasoning) and agentic AI (capable of autonomously executing multi-step diagnostic tasks). For instance, the AI agent developed by Ferber et al. achieved 87.5% tool-use accuracy and 91.0% correct clinical conclusions across 20 multimodal cases.
The goal of AI is not to replace pathologists but to augment their capabilities by handling routine tasks, allowing them to focus on complex diagnostic reasoning and clinical decision-making. Achieving this requires careful implementation, rigorous validation, and robust governance frameworks.
doi:10.2738/PR.2026.0005