Magnetic Resonance Imaging–Based Machine Learning Model in Classifying Jaw Lesions: A Preliminary Study

Objectives: This study assessed the efficacy of a demographic data and apparent diffusion coefficient (ADC) value-based machine learning (ML) model in the classification of odontogenic lesions. Methods: The dataset comprises magnetic resonance imaging data from 675 patients with various odontogenic lesions, including ameloblastomas, dentigerous cysts, and odontogenic keratocysts. The dataset includes demographic variables, ADC values, and types of lesions. Various models (random forest, logistic regression, support vector machine, extreme gradient boosting, and linear discriminant analysis) were trained and evaluated using five-fold cross-validation, augmented with synthetic samples. Learning curves and confusion matrices were used to assess the classification performance. Additionally, statistical analyses were performed to compare age, sex, and ADC values among the lesions. Results: The ML analysis revealed that the average recall, precision, and F1 score were all 0.87–0.95. Moreover, the median ADC revealed significant differences among the lesions (p < 0.001), as did sex (p = 0.04), although no age-based differences were observed. Conclusion: Our study demonstrated the efficacy of demographic data and the ADC value-based ML model for the classification of odontogenic lesions.