Related Experiment Video
Updated: Oct 19, 2026
Isolation of Cells with Morphological and Spatial Information from Oral Submucous Fibrosis Samples by Laser Capture Microdissection
Published on: August 11, 2023
Machine learning-based prediction of malignant transformation within 5 years in patients with oral leukoplakia: A
Qingmo Yang1, Mimi Zhou2, Chang Zhang3
1Department of Oral Medicine, Peking University School and Hospital of Stomatology, National Center for Stomatology, National Clinical Research Center for Oral Diseases, National Engineering Research Center of Oral, Biomaterials and Digital Medical Devices, Beijing Key Laboratory of Digital Stomatology, Research Center of Engineering and Technology for Computerized Dentistry Ministry of Health, NMPA Key Laboratory for Dental Materials, Beijing, China.
Background:
Oral leukoplakia (OLK), a potentially malignant oral disorder, has 7.2 - 9.8 % transformation rate. Identifying high-risk patients within 5 years can reduce the oral cancer burden.
Objective:
In this retrospective study, we aimed to identify the factors associated with malignant transformation in Chinese patients with OLK and develop and evaluate machine learning (ML) models for predicting malignant transformation within 5 years.
Methods:
In total, 672 OLK patients (203 showing transformation; 469 with no transformation) were included. Logistic Regression, SVM, Random Forest, GBDT, XGBoost, and LightGBM ML algorithms were used to develop models based on 17 clinicopathological features and evaluated using an internal test set and a temporal validation cohort of 73 patients (24 showing transformation; 49 with no transformation).
Results:
LightGBM showed the best overall performance in the internal test set (accuracy, 82.65 %; area under the curve (AUC), 0.8524), and it maintained good discriminative performance in the temporal validation cohort (AUC, 0.7959) when 17 clinicopathological features were used. XGBoost showed the best performance in both the internal test set (AUC, 0.8081) and temporal validation cohort (AUC, 0.8092) when five clinicopathological features were used. While predicting malignant transformation within 5 years, XGBoost achieved an AUC of 0.8505 in the internal test set. However, the temporal validation results should be interpreted with caution because of limited sample size.
Conclusion:
ML models may predict malignant transformation within 5 years in patients with OLK and may assist in individualized risk stratification. Further validation in larger multicenter cohorts is required before clinical application.
