Artificial Intelligence Driven Skin Cancer Detection Using R-FCN Enhanced Deep Convolutional Neural Networks with SMOTE Balancing

Skin cancer is a serious worldwide health issue, and earlier diagnosis is crucial for patient outcomes and efficient treatment. However, due to the variety of skin cancer types and the complexity of medical imaging, making an accurate diagnosis can be challenging. This study tackles this issue by introducing a new deep learning (DL) algorithm that is specifically designed for skin tumor diagnosis and employs the Convolutional Neural Network (CNN) technology. This study offers a novel approach that makes use of Region-based Fully Convolutional Networks (R-FCN) to address the crucial problem of skin cancer lesion categorization. The suggested system seeks to increase classification efficiency by using region-based detection which improves classification accuracy and localization. The HAM10000 and ISIC-2020 datasets, which are difficult and unbalanced, were used to thoroughly evaluate the created Deep CNN (DCNN) architecture. The Synthetic Minority Over-sampling Technique (SMOTE) was purposefully used as the method of random sampling in order to lessen the imbalanced datasets. This greatly enhanced the suggested models generalization and robustness. The results demonstrate the remarkable efficacy of the research contribution, yielding performance metrics consistently above 98% for F1-score, specificity, sensitivity, recall, accuracy, precision, and the area under the ROC curve (AUC). In terms of balancing speed and accuracy the suggested approach also performs better than traditional methods like R-CNN and YOLOv8. The study demonstrates that a strong framework for automatic skin cancer detection and classification is provided by combining R-FCN with SMOTE and CNN techniques. This framework facilitates early diagnosis and aids dermatologists in clinical decision-making.