Automated Skin Cancer Diagnosis: A Multi-Stage Deep Learning Approach for Malignancy Detection and Classification Using Dermoscopic Images

Tragically, skin cancer, a malignant skin disorder, is affecting an increasing number of individuals worldwide. Automatically diagnosing skin cancers using dermoscopy images as a follow-up to medical treatment methods is still a challenging undertaking. To address this issue, we provide deep learning models that can classify melanocytic tumors as either benign or malignant. There are four separate steps to the suggested cancer therapy protocol. As a preprocessing phase, dermoscopy images undergo a Laplacian-based hair removal approach before proceeding to noise reduction using a Median filter. It is also possible to get features from pre-processed photos. An method known as principal component analysis (PCA) may be used to extract features such as size, texture, color, and more. Dermoscopy images can accurately segment skin lesions because to these characteristics. Furthermore, the segmented skin tumor is categorized as either innocuous (non-melanoma) or hazardous (melanoma) using the ANU-Net technique. In order to segment the skin lesions, the LeNet-5 algorithm is used to find the lesion location and split them. It might be considered either aggressive or benign. By fixing these problems, the proposed technique achieves better accuracy and efficiency in cancer classification than competing methods. It is also difficult to manually detect early-stage cancer in dermoscopy images, which is another challenge. This paper proposes a multiple classification MSCDNet approach that utilizes dermoscopy photos to identify several cancer diagnoses. This study uses the stated MSCDNet method to categorize and segment various cancer types. Prior to processing, the dermoscopy images are corrected for distortion and noise. The BDA technique is used to choose desired attributes. The next step is to use the SqueezeNet model to categorize different skin lesions. Finally, dermoscopy image segmentation is performed using the DenseUnet model.