Skin Cancer Classification and Detection using Deep Learning based Techniques

Abstract

Skin cancer is a deadly type of malignant skin disease that has been increasing in worldwide incidence in recent years. As a follow-up to medical treatment strategies, automated identification of skin cancers using dermoscopic images has remained a challenging and difficult process. To overcome this issue, here, deep learning models were proposed to classify newlinethe melanocytic tumors as malignant or benign. The suggested approach for malignancy is categorized into 4 stages. In the Preprocessing stage the elimination of hair from dermoscopic images using a Laplacian-based method, then proceeds to noise removal using a Median filter. It is also feasible for extracting features from pre-processed pictures. The PCA method is used to extract features such as texture, dimension, and color. These characteristics aid in newlinethe proper segmentation of skin lesions from dermoscopic images. Furthermore, the segmen- newlinetation procedure employing LeNet-5 strategy for identifying the lesion location and dividing skin lesions, followed by classification; categorizing segment skin tumor as either innocuous newline(non-melanoma) or dangerous (melanoma) employing ANU-Net method. It is classified as either benign or aggressive. When compared to other options, the proposed technique resolves these difficulties in malignancy categorization and gives improved precision when classified. newlineIn addition, the difficulty of manually recognizing early-stage malignancy in dermoscopic images presents challenges. This study proposes a multiple classification MSCDNet approach employing dermoscopy images to identify several malignant diagnoses. Using the suggested MSCDNet approach, various malignancy disorders are classified and segmented in this study. newlineTo remove noise and distortion from the dermoscopic images, they are first pre-processed. newlineThe BDA method is used to select valuable attributes. Following that, the SqueezeNet model is used to classify different skin lesions. Finally, the dermoscopy images are segmented using the DenseUnet model. newlineMoreover, m