Enhancing Skin Lesion Analysis: Leveraging UNet and VGG Architectures in Deep Learning Models

Skin lesions encompass a broad spectrum of skin abnormalities, including benign moles, cysts, and various types of skin cancer. Skin lesion is amongst the most common group of cancer diagnosed worldwide with several types such as squamous cell carcinoma, melanoma and basal cell carcinoma. Factors such as prolonged sun exposure, genetics, and immune system status can contribute to the development of skin cancer. Timely detection of skin lesions leads to better treatment outcomes and decreasing illness severity. Identifying lesions in their initial stages allows for less invasive treatment options, preventing benign lesions from developing into cancerous forms. But the diagnosis and classification of skin lesions remains a tedious task and indeed requires a keen eye and expertise for precise classification. In this paper, we discuss how Computer-Aided Diagnosis (CAD) helps in the timely detection and analysis of these skin lesions, as well as the integration of technological advancements, specifically in computer vision and machine learning, into the field of dermatology. It aims to enhance diagnostic accuracy and treatment planning through the utilization of automated systems. The research addresses challenges encountered in this integration, such as handling unbalanced data and identifying precise areas of concern within dermatological images. Furthermore, the paper proposes the implementation of a model which resembles U-Net like architecture for precise segmentation of skin lesions, chosen for its ability to accurately delineate boundaries and capture crucial region for diagnosis. Additionally, a CNN model inspired by VGG architecture is proposed for lesion classification, leveraging its robustness in extracting complex features and patterns through deep and well-defined layers.