Intelligent skin cancer diagnosis using adaptive k-means segmentation and deep learning models

Since melanoma spreads swiftly throughout the body, it is typically a deadly form of skin cancer. Only when skin cancer is discovered early on is it usually treatable. In order to do this, this work proposes a unique melanoma detection model that has five main phases, including (i) pre-processing, (ii) segmentation, (iii) feature extraction, (iv) suggested HKPCA based dimensionality reduction, and (v) classification. Pre-processing is done first, and segmentation is done using a new adaptive k-means methodology after that. After that, features from the gray-level co-occurrence matrix (GLCM), deviation relevance based local binary pattern (DRLBP), and gray-level run-length matrix (GLRM) is extracted. Extracted features were subjected for dimensionality reduction via hybrid kernel proposed principal component analysis (HKPCA). These dimension reduced features are then classified using deep belief network (DBN) framework, where the weights will be optimized by means of improved elephant herding optimization (IEHO). Finally, a comparison of the proposed and existing models' convergent performance is conducted.