Pretreatment Identification of Oral Leukoplakia and Oral Erythroplakia Metastasis Using Deep Learning Neural Networks

Abstract

Without a doubt, Oral cancer is one of the malignancies worldwide which need to be diagnosed as early as possible because if not detected at early stage, the prognosis remains ineffective and can cause irreversible damage when diagnosed at advanced stages. Researchers have worked many years with Biopsy, Computerized Tomography (CT), and Magnetic Resonance Imaging (MRI) images for the precise identification. With the advancement of Medical Imaging, Machine Learning, and Deep Learning, early detection and stratification of oral cancer is possible. In this research, we have designed a Convolution Neural Network (CNN) model to classify oral cancer types: Leukoplakia and Erythroplakia on 550 oral images taken by the camera. We have trained our network with a Training-Validation ratio of 50–50%, 75–25%, and 80–20% on 20, 50, and 80 epochs. The comparative analysis has been performed using the precision, recall, f1-score, and confusion matrix. The highest accuracy achieved is of 83.54% with 0.87 f1-score for Leukoplakia and 0.78 f1-score for Erythroplakia. The proposed model accuracies were then compared with five different pre-defined architectures of CNN (VGG16, ResNet-50, Xception, EfficientNetB4, InceptionResNetV2).