A novel deep learning model for automated diagnosis of oral squamous cell carcinoma and related leukoplakia in pathological images.

BACKGROUND: Oral squamous cell carcinoma (OSCC) is the sixth most common type of cancer worldwide. Its diagnosis and treatment rely on pathological section analysis. However, the pathological diagnosis process is time-consuming, complicated, and may lead to errors due to human subjective factors. Therefore, the development of Automated and accurate auxiliary diagnostic tools based on deep learning are crucial. METHODS: This study proposed a novel deep learning model RRGNet, which was improved based on the RegNet architecture. By introducing the Ghost module and the residual channel attention module in the final stage of the model, and adopting strategies such as label smoothing, Mixup data augmentation, and SWALR learning rate adjustment, the feature extraction efficiency and computational cost were significantly optimized. The model was trained on a three-category dataset including OSCC and two types of vitiligo lesions, and the performance was compared with mainstream deep learning models such as GhostNet, HRNet, RegNet, ResNet50, and ViT . RESULTS: The experimental results show that the RRGNet model has an accuracy of 79.49% on the test set and an AUC value of >0.85, which is better than all the comparison models. At the same time, the number of parameters of RRGNet is significantly lower than that of other models, showing excellent computational efficiency and generalization ability. In addition, activation heat map analysis shows that the model can accurately focus on the lesion area and has strong clinical interpretability. CONCLUSION: The RRGNet model performed well in the three-classification task of OSCC pathological images, providing pathologists with an efficient and reliable auxiliary diagnostic tool. This study provides technical support for the early detection and precise treatment of OSCC. In the future, the universality of the model will be further verified on multi-center data, and a practical application system will be developed to promote clinical translation.