Initial Stage Oral Cancer Symptom Prediction of Tobacco and Gutka Consumers Using Deep Learning Method: A Case Study on India

In India, oral cancer stands as a significant public health challenge, predominantly fueled by widespread tobacco and gutka usage, necessitating innovative approaches for early detection.This study surpasses previous methodologies by harnessing cutting-edge deep learning techniques to detect subtle indicators of oral cancer, particularly crucial among tobacco and gutka consumers.By leveraging a vast dataset and employing transfer learning on pre-trained convolutional neural networks (CNNs), our model demonstrates enhanced proficiency in identifying early signs of oral cancer.To foster collaborative research, we curated a specialized dataset comprising 500 oral images of confirmed cancer cases and 400 images depicting non-cancerous oral conditions, made publicly accessible on Kaggle.Emphasizing the urgent need for targeted preventive measures, especially among individuals with a history of tobacco and gutka consumption, our study underscores the potential of deep learning technology to revolutionize healthcare interventions.The primary focus of our model lies in predicting early-stage symptoms of oral cancer, with significant implications for prognosis and treatment outcomes.Offering a non-invasive and efficient screening tool, particularly for high-risk groups like Indian tobacco and gutka consumers, our approach integrates deep learning seamlessly into the healthcare system.Through meticulous training and validation processes, we evaluate the efficacy of ResNet-50 and EfficientNetB0 models in classifying oral images as cancerous or non-cancerous.Employing augmentation techniques to enhance data diversity, we demonstrate robust performance metrics for both models.Results indicate commendable accuracy rates, with the EfficientNetB0 model exhibiting slightly superior performance over ResNet-50.The ResNet-50 model achieves a training accuracy of 98.66% and a validation accuracy of 95.79%, while the EfficientNetB0 model achieves a training accuracy of 99.44% and a validation accuracy of 96.32%.These outcomes underscore the potential of deep learning methodologies in enabling early detection of oral cancer, thereby offering invaluable insights for healthcare practitioners and researchers alike.Our study contributes to the growing body of knowledge on leveraging artificial intelligence for cancer diagnosis, particularly addressing public health challenges unique to the Indian subcontinent.By advancing the frontier of deep learning applications in healthcare, we pave the way for more effective strategies in combating the burden of oral cancer and improving patient outcomes.