Oral Disease Detection Using Advanced Hybrid Deep Learning Models

Early diagnosis of oral diseases, including oral cancer, improves survival rate, but most existing methods diagnose these diseases at a late stage, which hampers treatment efficacy. In this work, we present a hybrid deep learning model combining DenseNet-169 for feature extraction and LSTM for contextual analysis to improve diagnostic accuracy across a set of tongue conditions, including healthy tongue, hairy tongue, leukoplakia, oral thrush, oral lichen planus, and oral cancer. DenseNet-169 captures rich spatial features and LSTM processes sequential dependencies. This enhances the model's ability to capture fine disease patterns. Experimental results demonstrate better classification performance compared to state-of-the-art CNNs with higher accuracy, robustness and generalizability. This scalable, non-invasive approach improves early diagnosis with a reliable tool for clinical applications in oral health.