Artificial Intelligence in the Histopathological Diagnosis of Neoplasms: An Integrative Review of Diagnostic Accuracy and Clinical Efficacy (2019–2026)

Introduction: Conventional oncological pathology practice faces critical challenges stemming from interobserver variability and an ever-growing clinical workload. This review evaluates the technological maturity and clinical utility of artificial intelligence (AI) as a diagnostic and predictive support tool in histopathology. Methods: An integrative review of the literature (2019–2026) was conducted in PubMed, Scopus, and IEEE Xplore, following the methodology of Whittemore and Knafl. Studies on the diagnostic accuracy of deep learning algorithms in neoplasia histopathology were selected, with methodological quality assessed using QUADAS-2. Results: The evidence confirms that convolutional neural networks (CNNs) achieve diagnostic accuracy comparable to or exceeding that of pathologists in binary classification tasks, consistently reporting areas under the curve (AUC) > 0.98 in lung, breast, and prostate cancer. A disruptive finding is the validation of predictive computational histology, capable of inferring genotypic alterations—such as EGFR mutations or microsatellite instability—directly from standard hematoxylin and eosin (H&E) images, offering a cost-effective alternative for molecular screening. The evidence strongly supports the “augmented intelligence” model, in which the pathologist–AI synergy surpasses individual performance and mitigates visual fatigue. Conclusions: AI has transcended the experimental phase to become a robust technology for triage and digital phenotyping. Its definitive clinical adoption requires prioritizing multicenter external validation and the development of explainable AI (XAI) interfaces to overcome the “black box” barrier.