基于卷积神经网络的脑MR图像分割方法研究与应用

基于卷积神经网络的脑MR图像分割方法研究与应用(论文8000字)
摘要：因为利用核磁共振成像能轻松获得脑部的清晰图像，所以脑MR图像被广泛的运用到当今的医学研究和临床诊断中，帮助医生准确辨别病变组织和正常组织。然而在成像过程中产生的噪声和偏移场，使得传统的图像分割方法难以获得理想的效果。在目前常用的分割方法中，卷积神经网络因抗噪声效果明显和易于解决图像不均匀问题，所以被广泛用于脑MR图像分割。本文在此基础上，使用全卷积神经网络降低了训练过程的复杂程度，减少了分割时间，提高了分割精度，从而可以得到较好的分割结果。
关键词：核磁共振成像；图像分割；卷积神经网络；全卷积神经网络

Research and application of brain MR image segmentation method based on convolution neural network
School of Mathematics&Statistics，NUIST，Nanjing( 210044)
Abstract： It is easy to obtain the clear images of the brain by using MRI ,itssegmentedimages are widely used in medical research and clinical diagnosis to help doctors accurately identify diseased tissues and normal tissues. However, it is difficult to achieve the desired results with the conventional image segmentation methodsbecause of the noise and bias field during the imaging process . In now days, convolutional neural networks are widely used in MR image segmentation due to its obvious anti-noise effect and ease of solving the problem of image non-uniformity. In this paper, we use the full convolutional neural network to reduces the complexity of the training process, reduce the segmentation time and improve the segmentation accuracy, so that a better segmentation result can be obtained.
Key words：Magnetic Resonance Image; Image Segmentation;Convolutional Neural Network;Fully Convolutional Networks

目录
1.引言    1
2.卷积神经网络    2
2.1 卷积神经网络    2
2.2 卷积层    2
2.3 池化层    3
2.4 激活函数    4
2.5 分割原理    6
2.6 实验结果    6
2.7 结果分析    7
3.全卷积神经网络    7
3.1 全卷积神经网络    7
3.2 改造原理    7
3.3 Upsampling    8
4. U-net    9
4.1 网络结构    9
4.2 实验结果    10
4.3 结果分析    11
5.结论    11
参考文献    12
致谢    13