Next-generation AI framework for comprehensive oral leukoplakia evaluation and management

Area of Science:

• Oncology
• Artificial Intelligence
• Medical Imaging

Background:

• Oral potentially malignant disorders (OPMD) carry a significant risk of malignant transformation, especially with epithelial dysplasia (OED).
• Current diagnostic methods for OED are invasive and lack robust decision support for risk stratification and follow-up.
• There is a critical need for non-invasive tools to accurately assess OED and predict cancer risk.

Purpose of the Study:

• To develop and validate OMMT-PredNet, an automated multimodal deep learning framework for non-invasive OED identification.
• To enable time-dependent cancer risk prediction in patients with OPMD.
• To provide a reliable decision-support tool for optimizing patient management and improving oral cancer screening.

Main Methods:

• A multimodal deep learning framework, OMMT-PredNet, was developed using paired high-resolution clinical images and medical records.
• The model was trained and validated on a dataset of 649 histopathologically confirmed leukoplakia cases from multiple institutions.
• No manual region-of-interest (ROI) annotation was required, ensuring a fully automated workflow.

Main Results:

• OMMT-PredNet achieved high performance in cancer risk prediction (AUC: 0.9592) and OED identification (AUC: 0.9219).
• The model demonstrated excellent specificity and precision for both cancer risk prediction and OED identification.
• External validation confirmed the model's robustness and clinical applicability, supported by calibration and decision curve analyses.

Conclusions:

• OMMT-PredNet offers a non-invasive, automated solution for identifying OED and predicting cancer risk in oral potentially malignant disorders.
• The multidimensional deep learning approach effectively integrates clinical images and medical data for enhanced diagnostic accuracy.
• This framework has global applicability in improving oral cancer screening protocols and ultimately enhancing patient outcomes.