Skin Cancer Classification and Segmentation Using Deep Learning

Abstract

This paper integrates medical science and artificial intelligence, focusing on using convolutional neural networks (CNNs) to improve skin cancer diagnosis accuracy. Given the rising global incidence of skin cancers such as melanoma and basal cell carcinoma, this research is becoming increasingly important. This study uses the HAM10000 and PH2 datasets, which are known for their diverse skin cancer images, and employs a CNN-based approach informed by previous research findings.The proposed methodology includes extensive preprocessing and augmentation to increase the dataset's variability, allowing for thorough training and evaluation. The CNN model, which was developed using advanced training methods and includes convolutional and pooling layers, is the result of previous research demonstrating the efficacy of CNNs in skin lesion detection. Furthermore, the U-NET-based segmentation model contributes to the comprehensive analysis by precisely delineating lesion boundaries, which improves the understanding of skin cancer. The CNN model's performance is evaluated using a variety of metrics, including accuracy, classification reports, confusion matrices, and segmentation-specific metrics like the Dice coefficient and IOU. These metrics provide valuable insights into the changing landscape of skin cancer diagnosis, allowing for the development of effective, precise, and accessible healthcare solutions in the dynamic field of dermatology.