Related Experiment Video
Updated: Jun 28, 2026
Introduction of an Integrated Pathology Image Management, Artificial Intelligence, and Reporting System
Published on: July 11, 2025
Artificial Intelligence-Assisted Histopathologic Diagnosis and Grading of Oral Epithelial Dysplasia: A Systematic
Carlos M Ardila1,2, Eliana Pineda-Vélez3,4, Alejandro I Díaz-Laclaustra5
1Department of Periodontics, Saveetha Dental College, and Hospitals, Saveetha Institute of Medical and Technical Sciences, Saveetha University, Saveetha, Chennai, 600077, India. martin.ardila@udea.edu.co.
Purpose:
Artificial intelligence (AI) has emerged as a promising tool for digital pathology, with potential applications in the histopathologic evaluation of oral epithelial dysplasia (OED). This systematic review synthesized current evidence on AI-assisted histopathologic diagnosis and grading of OED, focusing on diagnostic performance, methodological quality, certainty of evidence, and functional roles of AI systems in the diagnostic workflow.
Methods:
A systematic search of PubMed/MEDLINE, Scopus, and Embase was conducted. Eligible studies evaluated AI, machine learning, deep learning, convolutional neural network (CNN), Transformer-based, or computational pathology methods applied to microscopic or digital histopathologic images of OED or oral potentially malignant disorders with dysplasia. Data on study characteristics, image format, analytical level, model architecture, reference standard, validation strategy, and diagnostic performance were extracted.
Results:
Thirteen studies published between 2017 and 2026 met the inclusion criteria. AI applications were grouped into three functional domains: detection of dysplastic epithelium, grading of dysplasia severity, and diagnostic assistance through enhanced image interpretation. Included studies used heterogeneous analytical approaches, ranging from handcrafted-feature machine-learning pipelines and CNN-based cellular or tissue-level classifiers to whole-slide image segmentation, DenseNet, Vision Transformer, and hybrid computational pathology pipelines. Several studies reported high experimental performance, including accuracy above 90% in selected detection or grading tasks. However, performance estimates were frequently derived from curated image-, patch-, region-, or slide-level datasets rather than patient-level diagnostic workflows. Substantial heterogeneity in datasets, grading frameworks, image formats, model architectures, validation strategies, and performance metrics precluded quantitative meta-analysis. Risk-of-bias concerns mainly involved patient selection, index-test evaluation, and flow/timing. The overall certainty of evidence was very low.
Conclusions:
AI shows potential to support histopathologic detection and grading of OED, but current evidence remains insufficient for clinical implementation. Future studies require standardized reference standards, transparent reporting, patient-level validation, external multicenter testing, and prospective evaluation in real diagnostic workflows.
