摘要
运用深度学习的方法基于脑部CT扫描图像合成相应的MRI。将28例患者进行颅脑CT和MRI扫描得到的CT和MRI的断层图像进行刚性配准,随机选取20例患者的图像输入U-Net卷积神经网络进行训练,利用训练好的网络对未参与训练的8例患者的CT图像进行预测,得到合成的MRI。研究结果显示:通过对合成的MRI进行定量分析,利用基于L2损失函数构建的U-Net网络合成MRI效果良好,平均绝对平均误差(MAE)为47.81,平均结构相似性指数(SSIM)为0.91。本研究表明可以利用深度学习方法对CT图像进行转换,获得合成MRI,现阶段可以达到扩充MRI医学图像数据库的目的,随着合成图像精度的提高,可以用于帮助诊断等临床应用。
The purpose of this research is to synthesize the corresponding magnetic resonance imaging(MRI)based on brain computed tomography(CT)images by deep learning.The tomographic images of CT and MRI obtained by brain CT and MRI scanning are rigidly registered in 28 patients,and the images of 20 patients are randomly input into U-Net convolutional neural network for training.The CT images of 8 patients who do not participate in the training are predicted by the trained network,thereby obtaining the synthetic MRI.The results reveal that through the quantitative analysis on synthetic MRI,the U-Net network constructed based on L2 loss function has a good performance in synthesizing MRI,with a mean absolute error of 47.81 and an average structural similarity index of 0.91.This study shows that deep learning method can be used to obtain synthetic MRI by converting CT images,thus achieving the purpose of expanding MRI medical database.With the improvement of the accuracy of image synthesis,it can be used in diagnosis and other clinical applications.
作者
董国亚
宋立明
李雅芬
李文
谢耀钦
DONG Guoya;SONG Liming;LI Yafen;LI Wen;XIE Yaoqin(State Key Laboratory of Reliability and Intelligence of Electrical Equipment,Hebei University of Technology,Tianjin 300132,China;Key Laboratory of Electromagnetic Field and Electrical Apparatus Reliability of Hebei Province,Hebei University of Technology,Tianjin 300132,China;Shenzhen Institutes of Advanced Technology,Chinese Academy of Sciences,Shenzhen 440305,China)
出处
《中国医学物理学杂志》
CSCD
2020年第10期1335-1339,共5页
Chinese Journal of Medical Physics
基金
深圳市配套项目(GJHS20170314155751703)
国家重点研发计划(2016YFC0105102)
国家自然科学基金(61871374)
广东省特支计划领军人才(2016TX03R139)
深圳市基础研究计划(JCYJ20170413162458312)
广东省自然科学基金(2017B20170413162458312,2015B020233011,2014A03-0312006)。
关键词
深度学习
CT
MRI
U-Net
卷积神经网络
图像模态转换
合成MRI
deep learning
computed tomography
magnetic resonance imaging
U-Net
convolutional neural network
crossmodality image synthesis
synthetic magnetic resonance imaging
