An Automatic Robust Deep Learning and Feature Fusion-based Classification Method for Early Diagnosis of Oral Cancer Using Lip and Tongue Images

Oral cancer is becoming a more challenging issue globally as it is 5
^th
most common cancer. Alcohol, betel nut and tobacco are responsible for more than 95% of oral cancer cases. Early diagnosis of oral cancer can improve survival prospects. Artificial intelligence is becoming more popular in medical diagnosis systems. This study aims to define the ability of AI models to analyze and identify early stages of oral cancer. Different deep-learning models were employed to develop an automated fusion-based network to recognize oral cancer. The proposed method is implemented on a publicly available dataset having tongue and lip images for the diagnosis of oral cancer. The data augmentation is applied in the dataset to avoid the problem of data unbalancing. The hyperparameters are optimally selected specifically for the considered dataset to provide higher accuracy and analyze the effectiveness of the proposed model. The evaluated performance of the proposed fusion-based model was compared with other state-of-the-art deep learning models which show that the proposed model outperforms all other models by achieving an overall accuracy of 94.62 %. The effectiveness of the proposed model can help in the medical diagnosis system for the detection and classification of oral cancer at an early stage.