Machine Learning for Predicting Malignant Transformation in Actinic Cheilitis: A Prognostic Support System Based on...

Area of Science:

• Oncology
• Dermatology
• Artificial Intelligence

Background:

• Actinic cheilitis (AC) is a precancerous condition with a risk of malignant transformation (MT).
• Accurate prediction of MT is crucial for timely intervention and improved patient outcomes.

Purpose of the Study:

• To develop and evaluate Machine Learning (ML) models for predicting MT in AC patients.
• To identify key clinical and demographic predictors of malignant transformation.

Main Methods:

• Utilized a dataset of 340 patients diagnosed with AC.
• Employed Adaptive Synthetic Sampling to balance the dataset for ML model training.
• Trained and validated four supervised ML classifiers (Random Forest, Xtreme Gradient Boosting, Multilayer Perceptron, Support Vector Machine) using 5-fold cross-validation.
• Applied SHAP values to determine the most influential predictors of MT.

Main Results:

• Xtreme Gradient Boosting (XGBoost) model demonstrated high performance with 96.72% accuracy, 96.87% sensitivity, and 0.9498 AUC.
• Multilayer Perceptron (MLP) achieved the highest sensitivity (98.44%), while Random Forest showed comparable results.
• Ulceration, multifocality, and long-standing lesions were identified as significant predictors of MT.
• Support Vector Machine (SVM) exhibited lower performance compared to other models.

Conclusions:

• XGBoost and MLP models show significant potential as decision support tools for monitoring AC patients.
• Synthetic data augmentation enhanced model robustness and predictive accuracy.
• The findings can aid clinicians in identifying high-risk AC lesions requiring closer surveillance.