Convoutional auto-encoder model-based cerebral blood flow impedance image reconstruction method
10.19745/j.1003-8868.2025174
- VernacularTitle:基于卷积自编码器模型的脑血流阻抗图像重建方法研究
- Author:
Xin-hui XU
1
;
Qiang DU
1
;
Li KE
1
Author Information
1. 沈阳工业大学,沈阳 110870
- Publication Type:Journal Article
- Keywords:
cerebral blood flow impedance imaging;
image reconstruction;
convolutional auto-encoder;
U-Net++network;
convolutional neural network
- From:
Chinese Medical Equipment Journal
2025;46(10):1-8
- CountryChina
- Language:Chinese
-
Abstract:
Objective To propose an image reconstruction method combining convolutinoal auto-encoder(CAE)and U-Net++network to solve the problems of ill-conditioned sensitivity matrix during cerebral blood flow impedance image reconstruction.Methods Firstly,the CAE model was optimized by introducing dense jump connections in the U-Net++network to enhance the perception of weak features in the sensitivity matrix.Secondly,the feature fusion mechanism in U-Net++network was combined to realize multi-scale fusion of the encoder and decoder in the CAE model,which improved the efficiency of feature transfer.Finally,the simulation data was used for pre-training and convolutional neural networks(CNN)was applied to predicting the conductivity,so as to implement high-precision image recontruction.In order to verify the effectiveness of the proposed method,the reconstruction results for five representative regions of blood flow changes were compared with those by the Tikhonov and conjugate gradient(CG)methods.Results When compared with the Tikhonov and CG methods,the proposed method had the average relative error decreased by 56.96%and 53.05%and the correlation coefficient increased by 19.37%and 5.79%for the reconstruction results of the five representative regions,respectively.The mean value of the structural similarity index by the proposed method was higher than 0.757 for the reconstruction results of the five regions,which was significantly higher than those by the other two methods.Conclusion The proposed method accurately reflects the size and location of blood flow changes in the brain region,enhances the precision and quality of image reconstruction and provides an effective solution for accurate reconstruction of cerebral blood flow impedance images.[Chinese Medical Equipment Journal,2025,46(10):1-8]
