| Abstrakt: |
Oral cavity cancer (OCC) is one of the most common cancers, in which more than 500,000 new cases and 350,000 related deaths occur every year worldwide. It is necessary to develop technologies that are objective, cutting-edge, and enable early, precise diagnosis. A patient's chances of surviving OCC are greatly increased when it is predicted in its early stages. In this work, a novel deep learning-based modified convolutional neural network (MCNN) for accurately detect and classify the normal, and abnormal cases of oral cavity. Initially, the images are gathered from Sheffield & Piracicaba datasets and these images are pre-processed using contrast limited adaptive histogram equalization which is used to reduce the noise from the input images. The proposed MCNN is improved version of convolutional neural network, in which the parameters of CNN are optimized with Stochastic gradient optimization algorithm to classify into normal and abnormal. The proposed MCNN is trained simultaneously using the initial dataset and the supplemented dataset. To train the MCNN, the source image size of the MCNN was used instead of resizing the complete images to fit the input file size of each pre-trained network. The effectiveness of the proposed MCNN is examined using several parameters like accuracy, precision, recall, specificity and F1 score. The experimental fallouts illustrate that the proposed MCNN achieves the overall accuracy of 97.96%, which is comparatively high than the state-of-the art techniques in OCC detection. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink