Abstract Background Artificial intelligence (AI) is increasingly applied in diagnostic pathology; however, synthesis of high-level evidence remains necessary to support clinical implementation. While numerous systematic reviews have evaluated AI performance, a high-level synthesis of this evidence is required to guide clinical adoption. Objective This umbrella review aims to synthesize evidence from existing systematic reviews and meta-analyses to evaluate the diagnostic accuracy of AI specifically within histopathology and cytopathology. Methods We conducted a comprehensive search for systematic reviews and meta-analyses employing AI in digital pathology. Studies focusing on non-pathological imaging (e.g., radiology, endoscopy) were excluded. Methodological quality was assessed using AMSTAR-2. Data regarding sensitivity, specificity, and AUC were synthesized qualitatively and, where strictly comparable, quantitatively. Results The review identified 6 systematic reviews covering key domains including prostate carcinoma, lymphoma, metastatic lymph node detection, and glioma grading. AI models, particularly Deep Learning (DL) algorithms, demonstrated high diagnostic performance, with sensitivity ranging from 83% to 96% and specificity from 78% to 95% across histopathological tasks. Performance was frequently comparable to expert pathologists in controlled settings. However, substantial heterogeneity was observed regarding scanning platforms, staining variations, and reference standards. Conclusion AI demonstrates strong potential as an assistive diagnostic tool in surgical pathology. However, current evidence is limited by the retrospective nature of primary studies and a lack of multicenter prospective validation. Future efforts must focus on standardized reporting and real-world clinical implementation trials.
Hyomshi Bhavsar (Wed,) studied this question.