MRA cerebrovascular image segmentation algorithm based on improved UNet
10.19745/j.1003-8868.2023198
- VernacularTitle:基于改进UNet的MRA脑血管图像分割方法研究
- Author:
Li MA
1
;
Yi-Fei SU
;
Zhen-Huan TAO
;
Wei-Dong YIN
;
Ying CHEN
Author Information
1. 南京市卫生信息中心,南京 210003
- Keywords:
cerebrovascular segmentation;
MRA image;
UNet;
residual neural network
- From:
Chinese Medical Equipment Journal
2023;44(10):7-12
- CountryChina
- Language:Chinese
-
Abstract:
Objective To propose a cerebrovascular image segmentation method for magnetic resonance angiography(MRA)based on improved UNet.Methods Firstly,the UNet network was used as the basic segmentation model and the residual neural network was incorporated to effectively alleviate the training pressure of the deep network and promote information transfer;secondly,the compression and excitation modules were added to improve the sensitivity of the network to cerebrovascular features;finally,the atrous spatial pyramidal pooling(ASPP)module was appended to obtain multi-scale feature information to further enhance the segmentation accuracy.The model based on improved UNet was tested on the TOF-MRA public dataset and compared with the models of UNet,ResNet and ResUNet++.Results The model based on improved UNet had a Dice similarity coefficient of 0.75 and an accuracy of 0.72,which were both higher than those of the models of UNet,ResNet and ResUNet++.Conclusion The method proposed segments MRA cerebrovascular images effectively,and thus can assist clinicians in disease diagnosis.[Chinese Medical Equipment Journal,2023,44(10):7-12]
