Abstract:

Abstract This paper presents a novel feature extraction model for accurate oral cancer detection using a combination of Modified Deep Convolutional Generative Adversarial Networks (MDCGAN) and Convolutional Neural Networks (CNN). The primary objective is to classify input Oral Cavity Squamous Cell Carcinoma (OCSCC) images as healthy or sick. The proposed approach involves image enhancement, where the input image is resized, contrast-enhanced, and converted from RGB to YCbCr color space using the Improved CLAHE method. The main novelty of this work lies in the deep learning-based feature extraction model, MDCGAN, which differs from traditional GANs in its use. In the proposed MDCGAN model, the Generator (G) part is employed to enhance the number of samples of each image in the dataset, thereby increasing the size of features and improving the accuracy of predictions. In contrast to conventional GANs, the Discriminator (D) part is replaced with a Modified Convolutional Neural Network (MCNN). The findings demonstrate that the proposed method outperforms existing approaches, achieving remarkable results during the testing phase with 97.26% classification accuracy, 98.96% precision, 94.18% recall, and 96.34% f-measure. The success of the oral cancer prediction depends on the quantity and quality of derived features from OCSCC images, making MDCGAN a highly recommended model for image classification applications compared to traditional deep learning approaches. In summary, the paper introduces a novel approach for oral cancer detection, combining MDCGAN for feature extraction and CNN for classification. The method showcases superior performance over existing techniques, emphasizing the importance of the derived features' size in achieving higher accuracy. The innovative use of GANs for feature extraction and MCNN as the Discriminator leads to improved oral cancer prediction accuracy, making MDCGAN an effective choice for such image classification tasks.

Keywords: