As specified by World Health Organization, the occurrence of skin cancer has been growing over the past decades. At present, 2 to 3 million nonmelanoma skin cancers and 132 000 melanoma skin cancers arise worldwide annually. The detection and classification of skin cancer in early stage of development allow patients to have proper diagnosis and treatment. The goal of this article is to present a novel deep learning internet of health and things (IoHT) driven framework for skin lesion classification in skin images using the concept of transfer learning. In proposed framework, automatic features are extracted from images using different pretrained architectures like VGG19, Inception V3, ResNet50, and SqueezeNet, which are fed into fully connected layer of convolutional neural network for classification of skin benign and malignant cells using dense and max pooling operation. In addition, the proposed system is fully integrated with an IoHT framework and can be used remotely to assist medical specialists in the diagnosis and treatment of skin cancer. It has been observed that performance metric evaluation of proposed framework outperformed other pretrained architectures in term of precision, recall, and accuracy in detection and classification of skin cancer from skin lesion images.

中文翻译：

---

一种基于健康事物驱动的深度学习框架，用于使用迁移学习检测和分类皮肤癌

正如世界卫生组织所指出的，皮肤癌的发生率在过去几十年中一直在增加。目前，全世界每年有 2 至 300 万非黑色素瘤皮肤癌和 132 000 例黑色素瘤皮肤癌。皮肤癌在早期发展阶段的检测和分类，可以让患者得到正确的诊断和治疗。本文的目的是提出一种新颖的深度学习健康与物联网 (IoHT) 驱动框架，使用迁移学习的概念对皮肤图像中的皮肤病变进行分类。在提议的框架中，使用不同的预训练架构（如 VGG19、Inception V3、ResNet50 和 SqueezeNet）从图像中提取自动特征，它们被馈入卷积神经网络的全连接层，使用密集和最大池化操作对皮肤良性和恶性细胞进行分类。此外，所提议的系统与 IoHT 框架完全集成，可远程用于协助医学专家诊断和治疗皮肤癌。已经观察到，在从皮肤病变图像中检测和分类皮肤癌的精度、召回率和准确性方面，所提出框架的性能指标评估优于其他预训练架构。