基于全卷积网络的皮肤镜图像上黑色素瘤分割
基于全卷积网络的皮肤镜图像上黑色素瘤分割(论文10000字,外文翻译)
摘要:随着深度学习在各行各业的渗透,医学领域发生了翻天覆地的变化,医疗效率和诊断准确率相应提高。皮肤癌中黑色素肿瘤是一种致命的癌症,要想精确对其分类,实现其病变区域分割是必不可少的。本文我们利用深度学习中的全卷积网络成功实现了皮肤镜图像上黑色素瘤的分割,并且达到了85.7%的平均精度和91.9%的总体精度。在此基础上我们进行了特征提取,并取得了一些成就。
关键词:黑色素瘤;全卷积网络;深度学习;分割
Melanoma Segmentation on Dermoscopy Images based on Fully Convolutional Networks
Abstract:With deep learning in all walks of life penetration, medical field has undergone enormous changes, medical efficiency and diagnostic accuracy correspondingly improved. Melanoma among skin cancer is a fatal cancer, in order to accurately classify it, achieving its lesion area segmentation is essential. In this paper, we use fully convolution networks in deep learning to successfully achieve the melanomasegmentation on dermoscopy images, and reach 85.7% of the average accuracy and 91.9% of the overall accuracy.On this basis, we have carried out the feature extraction, and has made some achievements. [资料来源:https://www.doc163.com]
Key words:melanoma;fully convolutionalnetworks;deeplearning;segmentation
[资料来源:http://www.doc163.com]
目 录
1绪论 1
1.1研究背景 1
1.2研究目的与意义 1
1.3国内外研究现状 1
1.4皮肤镜图像分割 2
1.5论文主要内容及章节安排 2
2深度学习与神经网络 3
2.1深度卷积神经网络 3
2.1.1背景 3
2.1.2卷积神经网络(CNN) 3
2.1.2.1输入变量 4
2.1.2.2特征 4
2.1.2.3滤波器(卷积核) 5
2.1.2.4感受野(receptive field) 6
2.1.2.5零填充 6
2.1.2.6超参数 7
2.1.2.7CNN结构 7
2.1.3全卷积网络(FCN) 9
3基于全卷积的黑色素瘤分割 11
3.1研究动机 11
3.2分割 11
3.2.1实验环境 11
3.2.2全卷积网络框架 12
3.2.3实验准备 13
3.2.4实验过程及结果 14
3.2.5实验分析 15
3.3 特征提取 16
3.3.1超像素 16
3.3.2特征 16
3.3.3预处理及训练 17
3.3.4结果分析 17
4结论与展望 18
4.1结论 18
4.2未来展望 18
参考文献 18
致谢 20 [资料来源:www.doc163.com]