Computer-Aided Diagnosis Of Skin Cancer From Dermoscopy Images

Abstract

The development of cancerous cells in skin tissues is a characteristic of skin cancer.  It poses severe health issues, and effective treatment relies on early detection.  This paper introduces an automated computer-aided approach for early detection of skin cancer.   The performance of a new image segmentation technique that utilizes a convolutional neural network optimized by satin bowerbird optimization (SBO) following image noise reduction is shown through a confusion matrix.  To obtain relevant information from the segmented images, feature extraction is subsequently performed.  In order to eliminate redundant data, an SBO algorithm-based optimal feature selection process is utilized. The pre-processed images are further split into two sets based on a support vector machine classifier: healthy or malignant.  The system is tested based on the performance indicators accuracy, sensitivity, negative predictive value, specificity, and positive predictive value in relation to American Cancer Society database through running the simulations and against 10 assorted techniques in the existing literature for comparison.