Multidimensional Artificial Intelligence-Based Cancer Progression Prediction in Oral Leukoplakia

Area of Science:

• Oncology
• Artificial Intelligence
• Biomarkers

Background:

• Oral cancer frequently arises from oral potentially malignant disorders (OPMDs).
• Oral leukoplakia (OL) is the most common OPMD, but cancer risk prediction remains challenging.
• Current risk stratification methods for OL have limitations.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI)-based time-to-event model, OralCancerPredict, for precise cancer progression prediction in OL.
• To integrate diverse data types including patient characteristics, histologic features, and immunohistochemical indicators.

Main Methods:

• Retrospective analysis of 456 patient data and 1,312 tissue samples from OL patients.
• Quantitative analysis of KRT13 and p53 immunohistochemistry combined with clinical and histopathologic data.
• Model training, testing, and external validation using independent datasets (119 patients, 322 tissue samples).

Main Results:

• OralCancerPredict demonstrated strong performance with high concordance indices (0.855-0.867) and AUCs (0.877-0.882).
• Explainability analysis highlighted the importance of KRT13/p53 expression and WHO dysplasia grading.
• The model showed superior net benefit compared to existing grading systems.

Conclusions:

• OralCancerPredict accurately predicts cancer risk and progression-free survival in OL patients.
• The explainable AI model offers improved risk stratification and clinical management potential.
• This tool can aid in streamlining interventions and close monitoring for OL patients.