Enhancing Melanoma Skin Cancer Detection and Classification: U-Net segmentation and feature fusion of Deep Learning Algorithms

Human skin cancer is the most prevalent type of cancer and poses a significant threat to life, particularly, melanoma skin cancer, which exhibits a high mortality rate. Accurate detection and classification of melanoma skin cancer continue to be essential for timely treatment and improving patient diagnosis. Traditionally, painful, invasive and time-consuming biopsies are used to detect melanoma skin cancer. However, recently, computer-aided diagnosis of melanoma has become crucial. Thus, in this study, the researcher proposes a novel approach combining the feature extraction capability of the VGG16 and ResNet-50 deep learning algorithms with the precise segmentation power of the U-Net architecture. By combining these methodologies, the study aimed to achieve enhanced accuracy in distinguishing melanoma skin cancer as benign or malignant. The proposed method leverages the U-net segmentation algorithm to accurately delineate lesion boundaries and highlight crucial diagnostic areas, and the VGG16 and ResNet-50 models to extract high-level features from melanoma images, capturing textures and intricate patterns indicative of melanoma skin cancer. With a precise segmentation algorithm and the fusion of rich features from deep learning models enable a comprehensive analysis of melanoma characteristics, facilitating more reliable classification. The study used a total of 10606 images of melanoma skin cancer, and the experimental result demonstrates promising results, achieving a classification accuracy of 93% in distinguishing benign and malignant cases. As the experimental result shows, it validates the effectiveness of the proposed method in accurate classification of melanoma skin cancer. This study contributes to advancing melanoma detection capability and paving the way for more accurate dermatological diagnostic tools