CT Skull Image Reconstruction Using Deep Learning Method Based on Magnetic Resonance Dixon Images:A Comparative Study
10.3969/j.issn.1005-5185.2025.04.018
- VernacularTitle:基于磁共振Dixon图像利用深度学习方法重建头颅CT图像的对比研究
- Author:
Hongfei ZHAO
1
;
Haipeng DONG
;
Qiong HUANG
;
Yuan QU
;
Keming LIU
;
Xiaomeng WU
;
Yurong SHANG
;
Xiping CHEN
Author Information
1. 上海交通大学医学院附属瑞金医院放射科,上海 200025
- Publication Type:Journal Article
- Keywords:
Convolutional neural network;
Magnetic resonance imaging;
Tomography,X-ray computed;
Deep learning;
Head;
Image reconstruction
- From:
Chinese Journal of Medical Imaging
2025;33(4):428-432,438
- CountryChina
- Language:Chinese
-
Abstract:
Purpose Based on a variety of combinations of cranial MR Dixon images,the deep learning method is used to generate CT images,and the reconstruction efficiency is evaluated by comparing with the corresponding CT images.Materials and Methods A total of 77 cranial CT and MR images were collected retrospectively in Ruijin Hospital,Shanghai Jiaotong University School of Medicine from June to December 2021.The U-Net neural network was used for network training,with 62 cases in the training set and 15 cases in the test set.CT image reconstruction was performed using four kinds of Dixon images and a total of seven models among the various combinations.Mean absolute error,mean squared error,Pearson correlation coefficient and skull area Dice similarity coefficient were used to evaluate the image reconstruction efficiency.Results The generated CT images of the various Dixon image combination models showed strong correlation with the corresponding CT images(R>0.75,P<0.05),and the CT images reconstructed by the four-channel model had the closest value to the actual CT images[mean absolute error=147.516±30.802,mean squared error=(8.648±3.403)×104],the highest correlation coefficient(R=0.796±0.055),and the highest similarity coefficient in the cranial region(Dice similarity coefficient=0.800±0.036).Conclusion Deep learning training through Dixon images can be used to generate CT images,and the combination of four kinds of Dixon contrast images can improve the CT image reconstruction efficiency.
