Predicting Progression of Oral Lesions to Malignancy Using Machine Learning.

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Oral lesions have a risk of malignant transformation.
• Early detection of oral cancer is crucial for patient survival.
• Predictive models can aid in identifying high-risk oral lesions.

Purpose of the Study:

• To develop machine learning models for predicting malignant transformation of oral lesions.
• To utilize large-scale electronic health record (EHR) data for predictive modeling.
• To identify key features associated with oral lesion progression to malignancy.

Main Methods:

• A retrospective cohort of 2192 patients with biopsied oral lesions was analyzed.
• Electronic health record data was used to extract demographic, clinical, and pathological variables.
• Machine learning models were trained and validated to predict malignant transformation.

Main Results:

• Machine learning models achieved an AUC of 86% for predicting malignant transformation in all oral lesions.
• For lesions with dysplasia, models predicted transformation with an AUC of 0.75.
• Key predictors included dysplasia grade, multiple lesions, anemia, and prior cancer history.

Conclusions:

• Machine learning models utilizing EHR data can accurately predict oral lesion malignancy.
• These models offer a feasible approach for identifying lesions likely to progress to cancer.
• Predictive analytics can support clinical decision-making in oral oncology.