基于深度学习算法的主动脉瘤CT影像分割技术研究

发布时间：2018-08-14 10:02

【摘要】：深度学习(Deep Learning)目前被广泛应用在很多科研领域,并且在工业生产中也得到了很多的应用,都取的了很多不错的效果。最近几年,深度学习算法的研究非常热门,已经有研究人员将深度学习应用到医学图像分析领域,深度学习算法的出现,使得计算机在处理大批量的图像数据变得更加容易,处理数据的速度更快。深度学习模型也越来越多,根据实现功能的不同,学习模型又分为很多种。比如说,深度卷积神经网络(Convolutional Neural Networks,CNNs)应用于图像分类,全卷积神经网络(Fully Convolutional Networks,FCN)应用于深度学习算法图像分割,不同的网络模型有不同的结构参数。医学图像分割就是将医学图像中病变区域给划分出来,将这些区域进行定量或定性的疾病分析,医学图像分割在医学图像处理过程中地位十分重要,是实现目标区域提取和定量表示的基础,能够为医疗诊断提供有用信息。在本文中我们将着重研究深度学习算法在主动脉瘤CT(电子计算机断层扫描技术)影像分割和OCT(光学相干断层扫描技术)脉络膜分割中的应用,提出的方法介绍如下:(1)为了解决能将CT影像中主动脉瘤分割出来,我们设计出了基于深度学习算法的TJ-1模型,利用我们已有的数据集,结合我们自己搭建的深度学习算法运行平台Caffe,实现对TJ-1分割网络模型的学习过程,最后通过实验的方式,来证明我们所设计的网络模型对主动脉瘤分割是有效的。(2)基于TJ-1模型,我们设计了TJ-2模型,利用TJ-2模型学习得到的OCT图像边缘权重,再结合图搜索算法来完成OCT脉络膜分割实验。通过实验,我们发现我们所提的方法,具有很好的应用前景,值得我们继续去深入研究。我们设计的实验方案,分割效果好,时间效率高。
[Abstract]:Deep learning (Deep Learning) has been widely used in many fields of scientific research, and has been applied in industrial production, and many good results have been obtained. In recent years, the research of depth learning algorithm is very popular. Some researchers have applied depth learning to the field of medical image analysis. The emergence of depth learning algorithm makes it easier for computers to process large quantities of image data. Data processing is faster. There are more and more deep learning models, which can be divided into many kinds according to the different functions. For example, the deep convolution neural network (Convolutional Neural networks) is applied to image classification, and the full convolution neural network (Fully Convolutional networks) is applied to image segmentation. Different network models have different structural parameters. Medical image segmentation is to divide the diseased areas in medical images and analyze these areas quantitatively or qualitatively. Medical image segmentation is very important in the process of medical image processing. It is the basis of target region extraction and quantitative representation, and can provide useful information for medical diagnosis. In this paper, we will focus on the application of depth learning algorithm in CT image segmentation and OCT (optical coherence tomography) choroidal segmentation of aortic aneurysms. The proposed methods are described as follows: (1) in order to solve the problem of segmenting aortic aneurysms from CT images, we design a TJ-1 model based on depth learning algorithm and use our existing data sets. The learning process of TJ-1 segmenting network model is realized by using our running platform of depth learning algorithm. Finally, the experimental method is used to prove that the network model designed by us is effective for aortic aneurysm segmentation. (2) based on TJ-1 model, the network model can be used to segment aortic aneurysm. We design the TJ-2 model and use the TJ-2 model to learn the edge weight of the OCT image and then combine the image search algorithm to complete the OCT choroid segmentation experiment. Through experiments, we find that the proposed method has a good prospect of application and is worthy of further study. The experimental scheme designed by us has good segmentation effect and high time efficiency.
【学位授予单位】：山东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP18

【参考文献】