This study presents a deep learning-based framework based on DenseNet169 for the classification of oral cancerous and non-cancerous conditions using lips and tongue images. In order to overcome the data scarcity and alleviate the overfitting, the Oral Cancer Images (OCI) dataset, consisting of 131 clinically validated samples, was increased to 1310 images using a ninefold augmentation strategy. Preprocessing – resizing, normalization, and augmentation to improve robustness were used. The DenseNet169 model showed an accuracy of 94%, with a precision, recall, and F1-score of 0.94, 0.95, and 0.95, respectively, to diagnose the cancerous cases, showing that the model is sufficient to diagnose the disease. Saliency-based visualization further enhanced interpretability by localizing lesion regions consistent with clinical observations. These results confirm the promise of DenseNet169 in supporting automated oral cancer screening, with potential to serve as an assistive tool in clinical decision-making and preventive oncology care.