AI-Based Histopathological Diagnosis of Ameloblastoma: Research Progress, Limitations and Prospects

Abstract

Locally aggressive and prone to recurrence, ameloblastoma (AME) is among the most common odontogenic epithelial tumors of the jaws. Histopathological diagnosis is often difficult because AME morphologically overlaps with other jaw lesions; where specialist expertise is scarce, misdiagnosis can lead to suboptimal surgery and poorer outcomes. This narrative review examines current evidence on AI-assisted histopathological diagnosis of AME. We describe the deep learning workflow and evaluate three tasks—pixel-level tumor epithelium segmentation, benign-malignant stratification, and multi-class differential diagnosis—with attention to intra-cohort comparisons. Reported results are encouraging: three-class classification reaches AUC 0.91, and tumor epithelium segmentation attains mean Intersection over Union (mIoU) 0.846, albeit in single-center retrospective settings. Clinical translation is hindered by cross-dataset domain shift, absent benchmarks, limited interpretability, and privacy constraints. We also outline an "AI screening plus pathologist review" arrangement that could improve efficiency. AI offers genuine promise as an adjunct, but progress will require standardized multi-center datasets, more interpretable algorithms, and prospective validation.