Automated Detection of Oropharyngeal Carcinoma Using Super-Resolution Generative Adversarial Network and Convolutional Neural Network

Oropharyngeal carcinoma is one of the prevalent forms of abnormal cancer. Detecting cancer at an early stage and getting proper treatment is the only way to reduce the mortality rate. The key intention of this work is to enhance the image quality and to classify the images with distinct test cases. The first approach is to implement an Enhanced Super-Resolution Generative Adversarial Network (ESRGAN) in the input stage for improving the quality of images. The second approach is to implement the U-Net architecture in a proposed method to enhance the accuracy rate and precise outcome for early prediction. After implementing a U-Net architecture, the cancerous image and non-cancerous images are identified and classified accurately with lesser computational time. The proposed methodology executes preprocessing of input images using de-gamma correction. Thereafter, the Convolutional Neural Network (CNN) model are employed to extract all the features pertained from the image which facilitates to classify the input images. The unique validation dataset consists of 1050 images utilized to analyze the proposed ESRGAN-CNN model. The simulated results have been tested under 2 test cases with different accuracy rates along with cancerous and non-cancerous categorized images. When compared to existing works, the suggested work produces a high-resolution image with superior accuracy of 92.51%.