SmartOralDx: A deep learning-powered system for precise classification of oral diseases from clinical imagery

The early and accurate diagnosis of oral diseases is essential for effective treatment and improved patient outcomes. This study introduces SmartOralDx, a deep learning-based diagnostic system designed to classify multiple oral disease categories from clinical imagery. The system was evaluated using various convolutional neural network (CNN) architectures, including baseline CNN, MobileNetV2, CNN + LSTM, and CNN + BiLSTM with Attention, across datasets comprising clinical and X-ray images. Initial results indicated that the inclusion of low-contrast X-ray images negatively impacted model performance. By refining the dataset to include only high-resolution clinical images and applying contrast-enhancement techniques using CLAHE, significant improvements were achieved in classification accuracy. The contrast-augmented CNN model achieved the highest testing accuracy of 94.26 %, while hybrid models incorporating temporal and attention mechanisms further enhanced interpretability and generalization, with the CNN + LSTM model reaching 90.75 % test accuracy. The study highlights the importance of data quality, augmentation, and model architecture in medical image classification and suggests that SmartOralDx has strong potential for integration into clinical workflows and mobile-based diagnostic tools. Future work will focus on expanding the dataset with diverse demographic inputs, deploying the system in real-time environments, and integrating it into smartphone-based platforms for broader accessibility. • Introduces SmartOralDx , a deep learning-powered diagnostic framework designed to classify multiple categories of oral diseases using high-resolution clinical imagery. • Evaluates and compares the performance of various deep learning models, including baseline CNN , MobileNetV2 , and hybrid architectures such as CNN-LSTM and CNN-BiLSTM with attention , on clinical and X-ray datasets. • Demonstrates that contrast enhancement using CLAHE significantly boosts model accuracy, with the CNN model trained on contrast-augmented clinical images. • Identifies that including low-quality X-ray images in the training set leads to performance degradation, underscoring the importance of image quality and preprocessing . • Highlights the potential of SmartOralDx for integration into mobile diagnostic applications and clinical workflows , supporting scalable, accessible early detection of oral diseases.