Automated skin lesion detection and classification using fused deep convolutional neural network on dermoscopic images

Abstract

AbstractSkin cancer becomes a deadly disease that affect people of all ages globally. The availability of various types of benign and malignant melanoma makes the skin lesion diagnostic process difficult. Since the visual inspection of skin cancer is costlier and lengthy process, it is needed to design automatic diagnosis model to classify skin lesions accurately and promptly. Computer‐aided diagnosis models can be employed to identify the presence of skin lesions using dermoscopic images. The automatic identification of skin lesions can assist the doctors and enable the detection process at an efficient and faster rate. With this motivation, this article presents an automated skin lesion detection and classification using fused deep convolutional neural network (ASDC‐FDCNN) on dermoscopic images. The ASDC‐FDCNN technique aims to identify the existence of skin lesions from dermoscopic images. The ASDC‐FDCNN model involves the design of two deep learning models namely VGG19 and ResNet152 models. Besides, the fusion based feature extraction process is performed to derive feature vectors. In addition, the DCNN technique was employed as classifier for identifying the presence or absence of skin lesions. The performance validation of the ASDC‐FDCNN technique takes place utilizing benchmark skin lesion dataset. A comparative results analysis reported the better performance of the ASDC‐FDCNN model over the recent technique with respect to various measures.