【作者】 张浩；

【导师】 杨华勇； 张斌；

【作者基本信息】 浙江大学 ， 机械电子工程， 2019， 硕士

【摘要】 近年来,采用计算机辅助诊断技术对医学相关图像进行分析和处理已得到越来越多的应用,尤其是对皮肤疾病相关医学图像的辅助诊断。皮肤疾病的诊断过程复杂,并且准确的诊断需要医生多年的经验积累,因此更加准确有效的图像辅助诊断方法对于皮肤病的及时诊治有很大帮助。传统的皮肤疾病辅助诊断技术需要人工设计提取特征,依靠这些特征进行诊断往往很难达到确诊要求,采用深度卷积神经网络的方式则可以减少人工干预,让识别模型自主学习特征,提高识别准确率。本文旨在研究基于深度学习的皮肤疾病图像分类和预测方法,并在所选数据集上对分类模型作一定程度的优化,提高模型对皮肤疾病的分类能力。主要工作概括如下:第一章主要分析介绍了采用深度学习的方式辅助诊断皮肤疾病的背景和意义,并在章节最后给出了本论文的主要内容以及结构安排。第二章分析了采用深度学习来识别皮肤疾病图像的框架,并建立了两个皮肤疾病数据集,分别是皮肤镜图像数据集和临床图像数据集,介绍了对应的预处理方法,包括图像去噪和图像增强两种方式。第三章主要分析了卷积神经网络的基本结构以及相关训练基础,并选定了相关的深度学习框架,最后在皮肤镜图像数据集上对比了不同卷积神经网络模型的识别效果,初步选定基础模型。第四章主要介绍了对上一章所选模型的改进,包括激活函数、池化方式、网络结构的改进,实验证明,本文所采用的改进方法确实在数据集上有更好的识别效果。第五章基于已训练完成的皮肤疾病识别模型建立了可视化界面识别系统,然后与医学相关人员进行了对比实验,证明了采用深度学习进行辅助诊断的研究意义。第六章对本文的研究进行了概括性的总结并给出对应结论,同时分析探讨了研究中存在的许多问题,给出意见和建议。更多还原

【Abstract】 In recent years,the use of computer-aided diagnostic techniques for the analysis and processing of medical-related images has been used more and more,especially for the auxiliary diagnosis of medical images related to skin diseases.The diagnosis process of skin diseases is complicated,and ’accurate diagnosis requires years of experience of doctors.Therefore,more accurate and effective image-assisted diagnosis methods can greatly help the timely diagnosis and treatment of skin diseases.Traditional dermatological disease-assisted diagnosis techniques require manual design of extraction features.It is often difficult to achieve diagnostic requirements by relying on these features for diagnosis.The use of deep convolutional neural networks can reduce manual intervention,allowing the recognition model to learn characteristics autonomously and improve recognition accuracy.This paper aims to study the classification and prediction methods of skin diseases based on deep learning,and optimize the classification model to a certain extent on the selected dataset to improve the classification ability of the model for skin diseases.The main work is summarized as follows:The first chapter mainly introduces the background and significance of using deep learning to assist in the diagnosis of skin diseases,and at the end of the chapter gives the main content and structure of this paper.The second chapter analyzes the framework of using deep learning to identify skin disease images,and establishes two skin disease datasets,which are dermoscopic image datasets and clinical image datasets,and introduces corresponding preprocessing methods,including images.Noise and image enhancement are two ways.The third chapter mainly analyzes the basic structure of the convolutional neural network and the related training foundation,and selects the relevant deep learning framework.Finally,the recognition effect of different convolutional neural network models is compared on the dermoscopic image dataset,and the base model is selected.The fourth chapter mainly introduces the improvement of the selected model in the previous chapter,including the activation function,the pooling method,and the improvement of the network structure.The experimental results show that the improved method adopted in this paper does have better recognition effect on the data set.The fifth chapter establishes a visual interface recognition system based on the trained skin disease recognition model.Then it is compared with medical related personnel to prove the research significance of using deep learning for auxiliary diagnosis.Chapter VI summarizes the research and gives corresponding conclusions.At the same time,it analyzes and discusses many problems in the research,and gives opinions and suggestions.更多还原