The skin, as the largest and one of the most vital organs of the human body, acts as a protective barrier between internal organs and the external environment. It performs several crucial functions, including protection, regulation, and sensation. When skin cells undergo genetic mutations, they can grow and multiply uncontrollably, leading to the formation of malignant tumors and, ultimately, the development of skin cancer. The primary cause of skin cancer is prolonged exposure to ultraviolet (UV) radiation from the sun or artificial sources like tanning beds, which damages the DNA in skin cells and promotes the growth of cancerous cells. Studies indicate that by the age of 70, one in five Americans will develop skin cancer, with more than two individuals succumbing to the disease every hour. However, early detection of skin cancer significantly improves the chances of successful treatment and recovery. Although numerous classification algorithms have been proposed in recent years to detect various stages of skin cancer, many still suffer from limited accuracy and high implementation complexity. In this project, we present a modified CNN-based skin cancer classification model utilizing six distinct architectures: VGG19, DenseNet201, InceptionV3, Xception, ResNet152, and MobileNetV2. These models were trained and evaluated using the HAM10000 dataset (Human Against Machine with 10,000 labeled training images). Our results demonstrate that DenseNet201 outperforms all other CNN architectures, achieving an accuracy of 97% along with the highest precision, recall, and F1-score. A comparative analysis of these models is also provided to highlight their performance differences.

Skin Cancer CNN Architectures and Classification Algorithms