Classification of Skin Lesions into Seven Classes Using Transfer Learning with AlexNet

Abstract

Melanoma is deadly skin cancer. There is a high similarity between different kinds of skin lesions, which lead to incorrect classification. Accurate classification of a skin lesion in its early stages saves human life. In this paper, a highly accurate method proposed for the skin lesion classification process. The proposed method utilized transfer learning with pre-trained AlexNet. The parameters of the original model used as initial values, where we randomly initialize the weights of the last three replaced layers. The proposed method was tested using the most recent public dataset, ISIC 2018. Based on the obtained results, we could say that the proposed method achieved a great success where it accurately classifies the skin lesions into seven classes. These classes are melanoma, melanocytic nevus, basal cell carcinoma, actinic keratosis, benign keratosis, dermatofibroma, and vascular lesion. The achieved percentages are 98.70%, 95.60%, 99.27%, and 95.06% for accuracy, sensitivity, specificity, and precision, respectively.

Skin Lesion Classification by Multi-View Filtered Transfer Learning

“…Esteva et al [24] demonstrated classification of skin lesions using a single convolutional neural network (CNN), trained end-to-end from images directly, using only pixels and disease labels as inputs, and it achieved performance on identification of the deadliest skin cancer and most common cancers; Gessert et al [25] proposed a patch-based attention architecture that provides global context between small and high-resolution patches with a novel diagnosis guided loss weighting method, outperformed previous methods and improves the mean sensitivity by 7%; Harangi et al [26] designed a deep convolutional neural network framework to classify dermoscopy images into seven classes, using GoogLeNet Inception-v3 and achieving remarkable improvement of 7%; Hosny et al [23] ing obstacles for the model's expansibility and efficiency, because it is necessary to convene professional doctors with costing much time to generate massive annotations. For the sake of relaxing this inconvenience, many exploitations [13], [14], [27]- [30] have been investigated to apply transfer learning into skin lesion classification. Specifically, transfer learning can distill useful knowledge from a source dataset to an unlabeled target domain, which means that it is unnecessary to mark cost-expensive annotations for target data only needing another existing dataset.…”

“…Specifically, transfer learning can distill useful knowledge from a source dataset to an unlabeled target domain, which means that it is unnecessary to mark cost-expensive annotations for target data only needing another existing dataset. Kessem et al [27] utilized a pre-trained GoogLeNet to conduct transfer learning on ISIC 2019, and it successfully classified the eight different classes of skin lesions; Hosny et al [28] presented an automatic skin lesion classification system with higher classification accuracy using the theory of transfer learning and the pre-trained deep neural network; Le et al [13] developed a deep learning system that can effectively and automatically classify skin lesions with an end-to-end deep learning process, transfer learning technique, utilizing multiple pre-trained models and novel class-weighted and focal loss, achieved top-1 classification accuracy of 93%; Mahbod et al [14] investigated and exploited the transfer learning-based skin lesion classification by a fusion approach with three-level ensemble strategy that exploits multiple fine-tuned networks; Alqudah et al [29] employed GoogleNet and AlexNet with transfer learning and gradient descent adaptive momentum learning rate for classification of skin lesion images; Hosny et al [30] proposed a highly accurate method utilized transfer learning with pre-trained AlexNet, and it achieved 98.70%, 95.60%, 99.27%, and 95.06% for accuracy, sensitivity, specificity, and precision, respectively.…”

Multi-Class Diagnosis of Skin Lesions Using the Fourier Spectral Information of Images on Additive Color Model by Artificial Neural Network

“…Also, the method proposed allows to classified a higher number of classes (i.e. ten) in comparison to [46], [47], but our proposal cannot detect images that do not belong to any of the ten classes, which is performed and reported in [46].…”

Target Recognition of Industrial Robots Using Machine Vision in 5G Environment

“…The NYU Depth V2 database is introduced (Chen et al, 2019) with the leave-one-out evaluation method adopted; that is, the dataset with a sample space of 1100 is divided into subsets of 1000 and 100, with the subset of 1000 used as the training set and subset of 100 as the test set. The constructed system model is compared with advanced CNNs (AlexNet, GoogleNet, LeNet, ZF-Net, and ResNet) (Wang et al, 2017;Fadlullah et al, 2018;Luo et al, 2019;Hosny et al, 2020;Wang and Jia, 2020). The following equation shows the accuracy.…”