Skin Cancer Diagnosis with AI: U-Net-Based Segmentation and Lite-CNN Classification

The diagnosis of skin cancer ranks as a prevalent deadly illness which requires prompt precise diagnosis methods for achieving needed therapy. The current medical practice for diagnosis depends on dermatological expert evaluation yet this method shows both time-consuming and subjective performance. A deep learning system comprised of U-Net for exact skin lesion segmentation and Lite-CNN for classification constitutes our proposed approach. The research dataset contains seven classification categories composed of Actinic Keratoses and Intraepithelial Carcinoma (AKIEC) along with Basal Cell Carcinoma (BCC) together with Benign Keratosislike Lesions (BKL) and Dermatofibroma (DF) and Melanoma (MEL) and Melanocytic Nevi (NV) and Vascular Lesions (VASC). The precise localization of affected skin areas happens because U-Net achieves a 96 % accuracy rate in lesion segmentations. The analysis process starts with segmented images that Lite-CNN classifies with 96 % accuracy for different target lesions. Evidence shows that this implementation framework provides physicians with a dependable and computational efficient tool for automatic skin cancer identification. Late diagnosis detection enabled by this system would help dermatologists make more accurate diagnoses thus leading to better patient results. Research will concentrate on optimizing the model speed and building a more extensive dataset for better model generalization.