基于2D深度学习网络的心脏MR图像分割
Cardiac MR image segmentation based on 2D deep learning network
投稿时间:2020-01-13  修订日期:2020-01-13
DOI:
中文关键词: 图像分割 全卷积神经网络 规范层 损失函数
英文关键词: image segmentation fully convolutional neural network batch normalization layer loss function
基金项目:
作者单位E-mail
谢勤岚 中南民族大学生物医学工程学院 xieqinlan@126.com 
张博 中南民族大学生物医学工程学院  
摘要点击次数: 74
全文下载次数: 0
中文摘要:
      针对手动和半自动分割方法在分割心脏图像中存在局限与不足的问题,采用基于全卷积神经网络的U-net网络来自动分割心脏左右心室及心肌。本文改进了U-net网络,在网络的上采样与下采样过程中加入规范层,同时选用权重交叉熵损失函数,从而加速网络收敛,降低网络过拟合,提高分割的精度。原始图像经预处理后加入改进的U-net网络中,得到训练模型,利用训练模型分割出心脏图像中的左心室、右心室和心肌。实验结果表明,该方法能够较好地分割出左心室等结构,克服了传统机器学习方法难以收敛的局限性。
英文摘要:
      Because of the limitations and shortcomings of manual and semi-automatic segmentation methods, it is difficult to extract the target region from cardiac images. U-net network based on fully convolutional neural network to automatically segment left and right ventricles and myocardium of the heart. This article improves the U-net network and the batch normalization layer is added in the network''s up-sampling and down-sampling processes, at the same time, weighted cross entropy loss function is selected, thus the batch normalization layer speeds up network convergence and reduces network overfitting and improves precision in segmentation. The original images are preprocessed and added to the improved U-net network, the training model is obtained and the training model is used to segment the left ventricle, right ventricle, and myocardium in the heart image. The experimental results show that this method can better segment the left ventricle and other structures, it effectively overcomes the limitation that traditional machine learning methods are difficult to converge.
View Fulltext   查看/发表评论  下载PDF阅读器
关闭