文档介绍:1
目录
摘要 2
1引言 5
课题背景和意义 5
CT成像基本原理 6
国内外研究现状及趋势 10
12
2 与研究相关卷积神经网络概要 14
与研究er, successfully made the first CT machine in 1967. X-ray computed tomography (CT) is widely used in clinical diagnosis because of its fast scanning time and clear imaging. In recent years, more and more attention has been paid to the problem of CT radiation. People use low-dose CT scan to reduce the harm of radiation to human body, but low-dose CT image will produce noise, which will lead to the decline of CT image quality, low-quality CT image will affect the doctor's diagnosis of the disease.
4
In order to improve the image quality of low-dose CT, this paper studies the image post-processing method of low-dose CT denoising based on deep learning convolution neural network. In view of the existing problems in the denoising method of low dose CT based on deep learning, two improvements are made. The main work is as follows:
(1) A convolution neural network denoising method for low dose CT data with calcification points is proposed. Because the calcification points in CT images are generally small and the data containing calcification points are few, the convolution neural network will remove the small calcification points as noise points. Firstly, a calcification adding method is designed to preprocess the training data. Then a low dose CT denoising network is designed. At last, two different loss functions are designed, which are used to train convolutional neural network in two steps. Compared with the current common methods, it is proved that the proposed method can retain the "calcification" and complete the low-dose CT denoising.
A deep learning convolutional neural network is proposed, which can be used to train unsupervised images of non aligned low dose CT and standard dose CT with different structures. At present, most low-dose CT denoising algorithms based on deep learning need a large number of low-dose CT images with the same str