Improved effect of image reconstruction algorithm on the basis of deep learning for automatic segmentation of ultralow dose CT on airway of children
10.3969/j.issn.1672-8270.2025.07.005
- VernacularTitle:基于深度学习的图像重建算法改善儿童超低剂量CT气道自动分割效果的研究
- Author:
Teng LU
1
;
Yun PENG
1
;
Haoyan LI
1
;
Hongwei TIAN
1
;
Yaoyao SONG
1
;
Jihang SUN
1
Author Information
1. 首都医科大学附属北京儿童医院 国家儿童医学中心影像中心 北京 100045
- Publication Type:Journal Article
- Keywords:
Ultralow dose CT;
Children;
Automatic segmentation of airway;
Image reconstruction;
Deep learning(DL)
- From:
China Medical Equipment
2025;22(7):25-29
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To evaluate whether the reconstructed image on the basis of deep learning(DL)can improve the success rate and display quality of automatic segmentation of computed tomography(CT)with ultralow dose for chest of children on airway.Methods:The clinical data of 41 consecutive cases who adopted ultralow dose CT to underwent reexamination on chest at Beijing Children's Hospital,Capital Medical University from February 2020 to September 2020 were selected,whose average age was(4.43±1.61 years).The scan protocol of ultralow dose CT was(0.05 mGy).The reconstructed images included 6 groups,which were respectively filtered reflection projection(FBP)image with 0.625 mm thickness,50%adaptive iterative recombination(ASIR-V)images,100%ASIR-V images,low energy DL(DL-L),medium energy DL(DL-M),and high energy DL(DL-H).The automatically segmentation software was used to conduct automatically segmentation for airway,and the success rate of automatic segmentation was recorded.For images that were successful segmented,a 5-point scale was adopted to subjectively evaluate the displayed quality for airway(5 point is the best).In addition,the CT values and noise values of the images of 6 groups for airway were objectively measured.Results:The success rate of automatic segmentation of DL-H image was the highest(60.98%),and that of the 100%ASIR-V was the lowest(39.02%).The subjective score of DL-H image of the automatic segmentation was the highest(4.06±0.55)point,and that of 100%ASIR-V was the lowest(2.44±0.76)point.DL-H can display more fine and small airways.The noise values of objective measurement showed that both of DL-H and 100%ASIR-V had the lowest noise value,and there was no statistical difference in that between them.Conclusion:The use of high energy deep learning iterative reconstruction(DLIR)algorithm can improve the success rate and display effect of automatic segmentation of ultralow dose CT for chest of children on airway,and DLIR is contribute to improve the accuracy of automatic segmentation algorithm of artificial intelligence.
