Skin Cancer Classification Detection using CNN and SVM

Abstract

Skin malignant growth is quite possibly the most commonly seen Malignancy type in people. Skin disease happens because of the un controllable developing of transformations occurring in DNAs developing to certain reasons. Perceiving the malignant growth in beginning phases could build the opportunity of an effective treatment. These days, PC helped finding applications are utilized nearly at each field. From the real dermo scopic images, the first-stage network aims for precise segmentation of the skin lesion. The second-stage network is a classification network that can predict the existence of Melanoma and Squamous Cell Carcinoma in a skin sample. Deep convolutional neural networks, such as Inception-v4, ResNet-152, and DenseNet-161, were trained for melanoma and squamous cell carcinoma detection and seborrheickeratosis classification. U-Net with VGG-16 Encoder was trained to create segmentation masks for lesion segmentation. Resnet engineering achieves the highest precision of 90 percent among the equations used in the proposed models.

Deep Residual Learning Image Recognition Model for Skin Cancer Disease Detection and Classification

“…In the classification task, the trained CNN attained an accuracy rate of 76.5 percent, a specificity of 89.5 percent and a sensitivity of 96.3 percent. Pushpalatha et al (2021) have employed CNN to identify skin cancer in the ISIC dataset by masking the tumour in the skin, which they claim is a novel technique. Their approach was a little different from the others in that they utilized 90 percent of the dataset for training and 10 percent of the dataset for data augmentation, in which they rotated the images by 10 degrees and zoomed them.…”

“…Skin cancer is diagnosed and classified using either classic machine learning (ML) or deep learning (DL) techniques (Pushpalatha et al, 2021). When using a typical ML approach, a domain expert must identify the applicable characteristics and make them more clearly apparent for the learning algorithm to function, hence reducing the complexity of the problem.…”