Automatic Early Skin Cancer Detection using MLBP and Regularized Extreme Learning

The incidence of skin cancer is increasing significantly, and the most common type of cutaneous tumour, basal cell carcinoma, is most effectively treated through excision. Post-surgery, squamous cell carcinoma may metastasise to distant sites if it infiltrates adjacent tissues or organs. Due to its rapid growth and dissemination, malignant melanoma accounts for 70% of fatalities associated with skin cancer. The prognosis of a patient is significantly improved by the timely diagnosis of skin cancer, especially malignant melanoma. The proposed approach encompasses preprocessing, segmentation, feature extraction, and model training to enhance early detection. Following the reduction of background noise in digital images using a median filter, Otsu's global thresholding method is employed for segmentation. The model is trained via RELM and MLBP. Subsequently, GLCM technology is employed to extract texture data. The method attains a success rate of 97.49%, surpassing the traditional MLBP and RELM algorithms. Its potential as an effective instrument for early skin cancer diagnosis is underscored by its remarkable accuracy. The proposed technology enhances dermatologists' diagnostic accuracy and speed by integrating advanced image processing with machine learning techniques. This will enhance therapeutic outcomes and survival rates.