Intelligent Skin Cancer Detection from Dermoscopic Images with Machine and Deep Learning Approaches

Abstract

Melanoma, in particular, is one of the most common and dangerous cancers in the world, and early diagnosis is critical to improving survival rates. A traditional diagnostic method, such as visual examination or dermoscopy, often requires expert intervention, but it can be challenging to distinguish early-stage melanoma from benign lesions. Artificial intelligence (AI), particularly machine learning and deep learning techniques, are applied to dermoscopic images in this study to detect skin cancer more accurately. Results showed that deep learning models were more accurate, more recallable, and had higher F scores than traditional machine learning algorithms. We compare the performance of Logistic Regression, K-Nearest Neighbors, and advanced deep learning architectures such as Xception, VGG16, and ResNet50 on two public datasets containing dermatoscopic images, HAM10000 and PH2. As a result of the study, deep learning models, especially when fine-tuned, offer significant improvements in detecting skin lesions, including melanoma, allowing for early detection.