Artificial Intelligence–Assisted Histopathological Assessment of Dysplastic and Premalignant Lesions Across Multiple Organ Systems
Abstract
Objective: Artificial Intelligence (AI) based digital pathology shows promise in enhancing the histopathological evaluation of dysplastic and premalignant lesions. However, conventional grading of oral epithelial dysplasia, cervical intraepithelial neoplasia, gastric intestinal metaplasia, adenoma, and ductal carcinoma in situ remains subjective and prone to inter-observer variation. This systematic review aimed to consolidate the evidence of the use of AI in the histopathological classification, grading, detection, and risk of progression of premalignant lesions in various systems. Data Sources: This systematic review followed PRISMA 2020 guidelines. Studies were searched from March 2019 to April 2026 in PubMed, Scopus, Web of Science and Google Scholar. Study Selection: Original English-language articles utilizing AI, machine learning, deep learning, whole-slide imaging, digital pathology or computer-aided diagnosis (CAD) for the classification of dysplastic or premalignant lesions using histopathological or histopathology-confirmed diagnosis were included. Data Extraction: Two reviewers independently extracted data. The risk of bias was evaluated using the modified QUADAS-2 and PROBAST, and the certainty of evidence was evaluated using GRADE. Data Synthesis: Twenty studies were included having data on oral, cervical, gastric, and breast lesions. AI models were found to be effective in the detection and grading of dysplasia, detection of high-risk lesions, assisting pathologists and prediction of malignant transformation. Conclusion: Premalignant lesion assessment can be aided by AI-driven digital pathology with a potential for enhancing reproducibility and clinical decision support. But, before it becomes a common practice, there is a need for prospective multicenter validation, standardized reporting, external testing and workflow-based clinical evaluation.