Digital Pathology for Cancer Diagnosis: Diagnostic Concordance With Conventional Microscopy: A Systematic Review and Metaanalysis

Background: Whole-slide imaging (WSI) has transformed digital pathology from a consultation and archival tool into apotential platform for primary cancer diagnosis. Because oncologic histopathology directly determines tumor classification,grading, staging, margin status, and treatment, digital interpretation must demonstrate clinically acceptable concordance withconventional light microscopy (CLM). The objective is to evaluate the diagnostic concordance of WSI-based digitalpathology with CLM for cancer and cancer-related histopathology and to synthesize clinically significant discordance, exactconcordance, intraobserver agreement, cancer-specific performance, and sources of disagreement. Materials and Methods:A systematic review and meta-analysis was structured according to PRISMA 2020. MEDLINE/PubMed, Embase, Scopus,Web of Science, and the Cochrane Library were searched using terms related to digital pathology, whole-slide imaging,cancer, conventional microscopy, diagnostic concordance, accuracy, validation, and non-inferiority. Primary studies directlycomparing pathologist interpretation using WSI and CLM in malignant, premalignant, cancer-screening, or oncologicallyrelevant histopathology were eligible. QUADAS-2 principles were used for quality assessment. Random-effects modelswere used for concordance outcomes, while non-inferiority and kappa-based results were synthesized separately whenoutcome definitions differed materially. Results: A total of 1,712 records were identified, 426 duplicates and 19preliminarily ineligible records were removed, and 1,267 records were screened. After 1,081 title/abstract exclusions, 186reports were sought, 179 full texts were assessed, and 146 were excluded. Thirty-three studies were included in thesystematic review, of which 28 contributed to quantitative synthesis. Twenty-five studies contributed to pooled clinicalconcordance, yielding a pooled estimate of 98.7% (95% CI 98.1–99.1%; I²=72.4%). Nineteen studies contributed to exactconcordance, with a pooled estimate of 94.1% (95% CI 92.2–95.6%; I²=84.6%). The pooled major-discordance proportionwas 1.7% (95% CI 1.1–2.5%; I²=64.8%). Intraobserver agreement was high, with pooled kappa approximately 0.89 (95% CI0.85–0.92). Subgroup concordance was highest in gastrointestinal and mixed multispecialty oncologic pathology and lowestin dermatopathology, reflecting greater variability in melanocytic and borderline lesions. Conclusion: Contemporary digitalpathology demonstrates very high clinical concordance with CLM for cancer diagnosis. Residual disagreement isconcentrated in dysplasia, nuclear atypia, grading, mitotic activity, small diagnostic foci, and melanocytic lesions rather thanin straightforward recognition of invasive malignancy. Properly validated WSI can therefore support routine primaryoncologic diagnosis when coupled with targeted quality assurance and local workflow validation.