Translating Artificial Intelligence in Pathology into Real Clinical Practice: NEJM AI | August 2026

Introduction:

Artificial intelligence (AI) has demonstrated remarkable ability to extract diagnostic, prognostic, and molecular information from routine hematoxylin and eosin (H&E)-stained pathology slides. Despite this rapid scientific progress, adoption of AI in pathology has been far slower than in radiology. This perspective examines the barriers preventing widespread clinical implementation and outlines practical strategies to translate pathology AI from research laboratories into routine patient care.

Why was this review needed?

AI models have shown impressive performance in digital pathology, but few have reached routine clinical practice.

Digital pathology infrastructure remains limited, with only a small proportion of pathology slides currently digitized.

Regulatory approval and clinical implementation of pathology AI significantly lag behind radiology.

Standardization, clinician trust, and governance remain major obstacles.

Practical implementation strategies are needed to realize the clinical benefits of AI.

Key Takeaways:

Despite rapid advances in AI research, clinical adoption of pathology AI remains limited, primarily because of infrastructure, workflow, regulatory, and economic challenges.

Digital pathology is still in its early stages, limiting the availability of high-quality whole-slide images required for AI deployment.

Variability in tissue preparation, staining quality, and scanning protocols continues to affect AI model performance and generalizability.

Building pathologist trust through transparent, explainable, and clinically validated AI systems is essential for widespread adoption.

Two major translational opportunities are emerging:

AI as a digital copilot, assisting pathologists with diagnosis, grading, biomarker assessment, quality assurance, and workflow efficiency.

AI-enabled clinical trials, improving patient selection, biomarker discovery, endpoint assessment, and trial efficiency.

Ensuring equitable access to AI technologies, particularly in low- and middle-income countries, will be critical to avoid widening global healthcare disparities.

Clinical Impact:

Rather than replacing pathologists, the future of AI lies in augmenting human expertise. AI-powered pathology has the potential to improve diagnostic accuracy, reduce reporting time, standardize interpretations, and uncover clinically relevant molecular information directly from routine H&E slides. Beyond routine diagnosis, AI is expected to accelerate precision oncology by enabling more efficient biomarker identification and smarter clinical trial design. Successful implementation will depend as much on digital infrastructure, validation, regulation, reimbursement, and clinician acceptance as on advances in AI algorithms themselves.

Bottom Line:

The greatest challenge in pathology AI is no longer algorithm development but successful clinical implementation. AI is poised to become a digital copilot for pathologists and a powerful enabler of precision medicine and clinical trials, provided healthcare systems invest in digital pathology infrastructure, rigorous validation, clinician trust, and equitable global access.