TransEns-Network: An Optimized Light-weight Transformer and Feature Fusion Based Approach of Deep Learning Models for the Classification of Oral Cancer
Karnika Dwivedi, Bharti Chugh, Anugrah Srivastava, Jai Prakash Pandey · International Journal on Computational Modelling Applications · 2024
Oral cancer is one of the most dangerous types of cancer that can threaten human health. The diagnosis of cancer and its possible disorder at an early stage is required to increase the survival rate. The transformer models are so popular in computer vision applications because of their ability to extract long-range relationships among data points. The combination of CNN and trans-former has attracted extensive study in classifying medical images on the limited dataset. In this work, a lightweight, fast and robust automatic transform-er-based network has been designed. The proposed network utilizes the capabilities of CNN models for feature extraction with transformer model to create a fusion of transformer and convolution model for the classification of oral cancer images. The transformer network can capture both the local and global dependence of the image features. The joint efforts of the transformer and CNN in extracting the features from images reduce the computation cost and complexity as well as increase the performance of the presented model. The performance of the model is tested on an unseen test set of a publicly available dataset of oral cancer. The result findings prove that the combined structure of CNN and transformer model can extract more discriminatory features which helps in improving the classification performance of the model. The comparative analysis with other state-of-the-art models determines that the proposed model achieved competitive performance among these considered models. In addition, the presented approach can be beneficial in the diagnosis system for the detection of oral cancer at an early stage as it is effective and can give more accurate predictions.