Integrating Explainable Artificial Intelligence into Histopathological Risk Assessment: A Scoping Review and Meta-analysis

Explainable artificial intelligence is increasingly used to support cancer detection, grading, and prognosis from histopathology, yet clinical adoption remains limited by uncertainty about reliability, interpretability, and governance. We conducted a scoping review and meta-analysis of peer-reviewed studies applying explainable AI to histopathology-based cancer risk assessment. Searches of PubMed, Google Scholar, Scopus, and Web of Science identified 47 eligible studies. Post hoc visual methods, particularly CAM and Grad-CAM, dominated the field. Nine studies contributed to meta-analysis, yielding a pooled area under the curve of 0.962 (95% CI 0.909–0.985) with substantial heterogeneity (I² = 97.1%). Subgroup analysis showed higher and more consistent performance at the slide level than at the patient level, identifying unit of analysis as a major source of heterogeneity. These findings support a translational roadmap for clinically meaningful explanation, usability testing, prospective validation, and governance-aligned deployment.