Automated Multiclass Skin Cancer Detection: An Efficient Approach Utilizing Pre-Processing and EfficientNet

Prevention of skin cancer is essential for enhancing treatment effectiveness and patient survival rates, which remains a fast growing worldwide health issue. This work proposed, a novel framework for multiclass skin cancer recognition that integrates advanced image pre-processing methods, feature extraction, and Deep Learning (DL) based classifications. The pre-processing stage involves resizing and normalization of dermoscopic images, which enhances uniformity and reduces noise, ensuring the images are suitable for subsequent analysis. For feature extraction, the Local Binary Pattern (LBP) is employed to capture critical texture information and patterns that differentiate between different types of skin lesions. Following feature extraction, an EfficientNet is used for the classification task. By combining the LBP's capability to capture fine-grained texture features with the robust learning capabilities of the EfficientNet, achieves high classification accuracy of 96.5%. The obtained results developed Python software and show that the integration of LBP with EfficientNet significantly improves detection accuracy, reduces false positives, and enhances the overall performance of the model.