Answers are grounded in this paper and cite their sources.
Histopathological study demonstrates machine learning accurately stratifies oral epithelial dysplasia using cellular morphometry, suggesting an objective adjunct to conventional grading.
Background: Histopathological grading of oral epithelial dysplasia (OED) remains an important component of riskassessment in oral potentially malignant disorders. However, conventional assessment relies on qualitative interpretation ofarchitectural and cytological abnormalities and may be affected by observer variability. Quantitative computationalmorphometry offers an opportunity to convert microscopic morphological characteristics into reproducible numericalvariables that can subsequently be analyzed using machine-learning algorithms. Materials and Methods: A total of 81histopathologically evaluated specimens were included, comprising 27 normal buccal mucosa (NBM), 27 low-risk dysplasia(LR), and 27 high-risk dysplasia (HR) cases. Digital photomicrographs were subjected to automated computational analysisusing QuPath. Nuclear and cellular parameters were extracted, including nuclear area, nuclear perimeter, nuclear circularity,cellular area, cellular perimeter, cellular circularity, and nuclear-to-cytoplasmic ratio. The extracted features weresubsequently used for classification using Logistic Regression, Support Vector Machine, Random Forest, Decision Tree, KNearestNeighbors, and Gaussian Naïve Bayes. Classification performance was assessed using accuracy, precision, recall,and F1-score. Results: Progressive morphological alterations were observed across NBM, LR, and HR groups. Nuclear areaand nuclear perimeter increased with increasing dysplastic severity, whereas nuclear circularity decreased. Cellular areademonstrated a progressive reduction, while cellular perimeter increased. The nuclear-to-cytoplasmic ratio increasedprogressively across the three groups. Differences in the evaluated morphometric parameters were statistically significant.Among the evaluated machine-learning models, Random Forest demonstrated the highest overall classification accuracy(86.73%), followed by Decision Tree (78.96%), Support Vector Machine (68.42%), Logistic Regression (64.38%), KNearestNeighbors (59.84%), and Gaussian Naïve Bayes (52.47%). Conclusion: Quantitative morphometric characteristicsdemonstrate systematic alterations associated with increasing severity of oral epithelial dysplasia. The findings indicate thatcomputational morphometry can provide objective numerical descriptors of dysplastic morphology and that machinelearningalgorithms, particularly Random Forest, can utilize these features for automated classification. Computationalmorphometry combined with machine learning may therefore serve as a useful adjunct to conventional histopathologicalassessment. Larger, multicentric datasets and independent external validation are required before clinical implementation.
No takes yet. Share an insight, caveat, or question.
Kuppusamy et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: