OralNet: deep learning fusion for oral cancer identification from lips and tongue images using stochastic gradient based logistic regression

Timely detection of oral cancer plays a critical role in improving survival rates. While traditional biopsy procedures can be invasive and uncomfortable, a more non-intrusive and convenient alternative is fluorescence visualization using optical instruments. This method not only provides real-time results but also facilitates repeat examinations. In the current research, an innovative strategy for oral cancer identification is introduced, utilizing images of lips and tongue. This involves the Feature Fusion Deep Convolution Neural Network with Stochastic Gradient based Logistic Regression for the diagnosis of Oral cancer from lips and tongue images. A benchmark dataset is assembled, and the ReNet-50 model is utilized to extract multilayer convolutional features. Subsequent layers, including pooling, transformation, and fusion layers, are designed to handle hierarchical features across different branches. Finally, the proposed model undergoes training via logistic regression on the extracted data using the Cross Entropy Loss, and the optimizer (Stochastic Gradient Descent with weight decay) is employed to update the model parameters. This type of integration provides a powerful tool for leveraging deep learning capabilities while maintaining interpretability through the logistic regression layer. The outcomes from the experimentation on the Oral Cancer (Lips and Tongue) Images Dataset demonstrate highly competitive classification results, achieving an accuracy of 97.8%, which compares favourably with numerous methods. This research significantly contributes to ongoing endeavours aimed at developing non-invasive and efficient techniques for early oral cancer detection, potentially enhancing patient outcomes and reducing mortality.