13 - Neutrosophic multiple deep convolutional neural network for skin dermoscopic image classification

Abstract

Skin cancer is a fatal worldwide disease that has complicated types of different shapes, colors, patterns, and characteristics. Accurate and early diagnosis of skin lesions is a complicated task owing to the fuzzy characteristics of skin cancers. Lately, deep-learning models, including convolution neural networks (CNNs), have the ability to break through the analysis and classification processes in several medical applications. This chapter shows an automated computer-aided diagnosis (CAD) system for classifying skin lesions in dermoscopy images using deep learning networks. The proposed classification model seamlessly incorporates the proposed multiple deep convolution neural network (MDCNN) with a Neutrosophic Similarity Score (NSS) procedure to form a two-stage framework. In the proposed neutrosophic multiple deep convolution neural network (NMDCNN) model, the NSS is used to determine the reinforced training number for each epoch until all epochs are finished during the training process for each DCNN model. Furthermore, the incremental learning strategy and the maximum voting scheme are employed in the multiple deep convolution neural network to alleviate the final classification of the dermoscopic images into two classes: malignant or benign. The proposed framework is substantially assessed on the public International Skin Imaging Collaboration (ISIC) dataset. Evaluation metrics of the proposed NMDCNN proved the competency of the proposed model.