Classification of Oral Potentially Malignant Disorders Using Multimodal Feature Integration

Early detection of Oral potentially malignant disorders (OP-MDs) is vital for improving patient outcomes, especially in regions with limited access to advanced diagnostic facilities. The study proposes a new classification framework for early detection of Oral potentially malignant disorders (OPMDs), which combines RGB and hyperspectral features. Traditional deep learning models require extensive datasets, which are often unavailable in medical imaging due to privacy concerns and the rarity of certain conditions. The framework uses the Multi-stage Spectral-wise Transformer (MST++) for spectral reconstruction, enriching RGB images with detailed spectral information without expensive equipment. The framework extracts a comprehensive set of features, including Hu Moments, Haar texture features, SIFT descriptors with Bag of Visual Words, RGB statistical features, and hyperspectral spectral features, to form a rich representation of the data. The optimized random forest algorithm is used for classification. The approach shows effectiveness in scenarios with limited and imbalanced data, offering a practical solution for early detection of OPMD in resource-constrained settings. Codes are available in the GitHub repository.