Exploring the Use of Neural Networks for Early Detection of Oral Cancer and Other Dental Pathologies

Delayed diagnosis and restricted access to specialist evaluations are significant issues in the detection of oral cancer and other dental conditions, resulting in unfavorable treatment results. This work investigates the possibilities of neural networks for early diagnosis by comparing five deep learning algorithms: CNN, ResNet, InceptionV3, DenseNet, and a hybrid algorithm (InResNet). The models were experimented with and validated with a labeled image dataset of oral pathology, validating their accuracy, sensitivity, and specificity. Results of experiments are presented that reflect InResNet’s highest accuracy (96.7%), sensitivity (95.8%), and specificity (97.1%), surpassing those of CNN (92.5% accuracy), ResNet (94.3%), InceptionV3 (93.8%), DenseNet (95.1%), and VGG16 (90.2%). Results indicate the best feature extraction and classification performance by InResNet for early diagnosis. The research underscores the significance of sophisticated neural networks in enhancing diagnostic accuracy, but additional studies must be conducted to optimize models for real-time use, dataset variety, and integration with clinical practice.