Temporal Analysis of Oral Lesions Using RNN for Early Detection of Oral Cancer

Oral cancer represents a severe and prevalent health issue, especially in low-and middle-income nations. Timely diagnosis is essential for enhancing survival rates; however, conventional diagnostic methods tend to be costly and invasive. A self-operated oral cancer detection system makes use of Recurrent Neural Networks (RNNs) optimized through Adam and RMSprop techniques. The training and testing dataset comprises a Histopathologic Image dataset sourced from Kaggle, featuring 2,494 images classified as normal and 2,698 images indicative of oral cancer. Furthermore, real-time images, including 44 normal and 87 cancer-affected samples, were employed to assess the model’s applicability in practical scenarios. The system demonstrates an accuracy level of 97.41% which indicates its potential to boost diagnostic reliability. The main role of RNNs in oral cancer detection results from their ability to handle sequential data thus enabling precise recognition of spatial-temporal patterns across histopathologic images and dynamic lesion changes. By utilizing the Adam and RMSprop optimization techniques, the model’s performance and convergence are significantly im-proved, leading to consistent and reliable outcomes. This study emphasizes the necessity of implementing advanced deep learning techniques for non-invasive, cost effective, and efficient cancer detection. Future investigations will focus on expanding the methodology to encompass larger datasets and further refining the detection of oral features to enhance diagnostic accuracy.