To explain why high diagnostic accuracy alone does not establish the clinical value of AI in digital pathology and to propose a framework for evaluating implementation.
Approach:
Diagnostic pathway framework: The authors link six stages: tissue preparation, image acquisition, AI model performance, laboratory workflow, clinical decision-making, and patient outcomes.
Implementation assessment: The framework calls for external validation in representative populations, local testing, workflow assessment, and ongoing performance monitoring.
Key Findings:
Differences in tissue preparation, scanners, and laboratory procedures can affect image quality and AI performance; results from one laboratory may not transfer to another.
In research involving 28 pathology experts, incorrect AI recommendations overturned previously correct assessments in approximately 7% of consultations.
The authors argue that evaluation should establish whether AI integrates safely into workflows, supports appropriate decisions, and produces measurable patient benefits.
Interpretation:
The editorial argues that technical accuracy does not necessarily translate into better clinical judgment or improved patient outcomes.
Limitations:
The proposed framework is conceptual and has not been validated in prospective studies.
Conclusion:
The authors present the framework as a structured way for laboratories to identify evidence gaps and assess whether pathology AI is ready for clinical use.
