2021 1st International Conference on Emerging Smart Technologies and Applications (eSmarTA) | 2021
Skin Lesions Recognition System Using Various Pre-trained Models
Abstract
Globally, skin lesion is known as one of the deadliest diseases among humans. Such that skin cancer, which is recognized as dangerous cancer, causes death in many cases. Therefore, many scholars have investigated in this area to design automated skin lesion recognition systems. The variability of the skin diseases appearance makes the diagnostic task very difficult. This paper presents an image-based diagnosis system using convolutional neural networks (CNNs) to exploit an automatic recognition model for three common skin lesions (Melanoma, Nevus, Benign Keratosis Lesion BKL) using dermoscopic images. In this study, eight pre-trained models have been implemented ResNet18, ResNet-50, ResNet101, VGG11, DenseNet121, InceptionResNetV2, AlexNet, and GoogLeNet. Different classifiers are used including neural networks and a multilevel fine-tuning method. The experiments were carried out based on a new version of the ISIC dataset. In terms of accuracy, the best results among the implemented models were achieved by Inception_ResNet-V2, which reached 0.875 and 0.87 in the training phase and testing phase respectively.