Oral Cancer Identification Model using improved deep transfer learning based optimized framework

Oral cancer is a major health problem globally, particularly in rural and resource-poor areas, where mortality is increasing due to late detection. In this paper, an automated deep learning-based framework for clinical image based oral cancer detection is proposed. The approach combines Gaussian filtering and image preprocessing, U2- Net-based lesion segmentation, deep feature extraction using Efficient Net-B0 and an enhanced deep transfer learning (IDTL) classifier optimized with the Leaf in Wind Optimization (LiWO) algorithm. The developed method makes it possible to distinguish malignant oral lesions from those ones that are not. The suggested method outperforms conventional machine learning techniques and presenting the experimental results using a publicly available Kaggle dataset, highlighting its suitability for reliable clinical and rural screening applications. Additionally, the light weight architecture and optimized learning strategy ensure computing efficiency making the system suitable for real-time deployment. The formulated framework has strong potential to support early diagnosis and improve oral cancer outcomes in limited-resource healthcare systems.