Deep Learning Based Melanoma Diagnosis Identification

Abstract Malignant melanoma is considered to be one of the deadliest types skin cancer, and it responsible for death a large number people worldwide. However, distinguishing whether benign or malignant has been challenging task. Many Computer Aided Diagnosis Detection Systems have developed in past this This paper presents deep learning framework based approach diagnosis recognition. In proposed method, original mirror image first preprocessed then passed VGG16 convolutional neural network tumor property classification. uses smaller kernels instead larger kernel achieve reduction parameters thus improve performance. The system trained using segmented RGB images generated from ground truth ISIC2016 dataset, finally softmax classifier used pixel-level classification lesions. study, new method become lesion was designed classify regions into tumors on results classification, experiments were conducted two well-established public test datasets, ISIC2017, with final accuracy 96.1%. indicate that networks are suitable identification. study great relevance advanced cancer caused by melanoma.