INCS: Design and Development of an Oral Cancer Identification Methodology based on Improved Neural Classification Scheme

The early detection of oral cancer is critical to improving patient outcomes and survival rates. This paper presents the Improved Neural Classification Scheme (INCS) for identifying oral cancer using deep learning techniques. The proposed model integrates MobileNet and EfficientNet, two state-of-the-art convolutional neural networks (CNN), to create an ensemble capable of extracting both fine-grained and high-level semantic features from oral cancer images. The dataset used consists of 500 oral cancer images and 450 non-cancerous images sourced from Kaggle. A comprehensive preprocessing pipeline, including resizing, normalization, and data augmentation, was implemented to optimize the dataset for model training. The INCS model achieved an accuracy of 98.65%, outperforming other models such as MobileNet, EfficientNet, ResNet, and VGG16. The model also achieved high scores in precision (96.89%), recall (97.45%), and F1score (97.17%), demonstrating its robustness in distinguishing between cancerous and non-cancerous lesions. This high accuracy, combined with its low false positive and false negative rates, highlights the potential of INCS for clinical application in oral cancer diagnosis. Future work will focus on expanding the dataset and exploring real-time deployment in clinical settings.