Skin Cancer Classification using A Hybrid Convolutional Neural Network with SVM Classifier

Digital dermoscopy is commonly employed by dermatologists for the diagnosis of malignancy. Clinicians' recognition of tumor is not only highly subjective, but also highly dependent on their level of experience. The inherent inter-operator unevenness in the private processing of dermoscopy pictures calls for fully automated computer-aided diagnosis(CAD) methods. The fine-grained differences in the manifestation of these abrasions on the skin-surface make automated skin lesion classification difficult. In several different types of fine-grained object classifications, deep convolutional-neural-network(CNN) has demonstrated excellent reparability. In order to categorize dermoscopy pictures into benign or malignancy lesions, this research introduces two innovative hybrid CNN strategies that use a support-vector-machine (SVM) at the output layer. In order to classify data, the characteristics mined by the first &second CNN strategies are combined and supplied to the SVM. The suggested model's efficacy is measured against a reference set of labels provided from a dermatology specialist. In a study using CAD for skin disease categorization, the generated CNN-segmented image folder distinguished atopic dermatitis from nasty diseases and obstacles like mycosis fungoides, impetigo, and herpesvirus contagion with a sensitivity and specificity of around 98%. In comparison to the traditional CNN strategies, the proposed models performed better on the ISBI 2016 public dataset. Both of the proposed models, with an accuracy of 88.02% and 87.43%, outperformed the standard CNN models. Reduced reliance on physician annotation and enhanced CAD performance are both benefits of our CNN segmentation algorithm, which automatically removes lesions and segments images of the skin area across all image fields.