Deep learning for size and microscope feature extraction and classification in Oral Cancer: enhanced convolution neural network

Background and Aim: Deep learning technology has not been implemented successfully in oral cancer images classification due to the overfitting problem. Due to the network arrangement and lack of proper data set for training, the network might not produce the required feature map with dimension reduction which result in overfitting problems. This research aims to reduce the overfitting by producing the required feature map with dimension reduction through using Convolutional Neural Network. Methodology: The proposed system uses the Enhanced Convolutional Neural Network and the autoencoder technique to increase the efficiency of feature extraction process and compresses the information. In this technique, unpooling and deconvolution is done to generate the input data to minimize the difference between input and output data. Furthermore, it extracts characteristic features from the input data set which regenerates the input data from those features by learning a network to reduce the overfitting problem. Results: Different value of accuracy and processing time is achieved using different sample group of Confocal Laser Endomicroscopy (CLE) images. Based on result, it shows that the proposed solution is better than the current system. Also, the proposed system has improved the classification accuracy by 5 ~ 5.5% in average and reduced the processing time by 20 ~ 30 milliseconds in average. Conclusion: The proposed system is focused on accurately classifying the oral cancer cells of different anatomical locations from the CLE images. Finally, this study enhances the accuracy and processing time using autoencoder method and solve the problem of overfitting.