Influence of image reconstruction algorithms on abdominal portal-phase CT histogram and wavelet features in patients with hepatic tumor
10.3760/cma.j.cn112149-20241124-00697
- VernacularTitle:不同重建算法对肝脏肿瘤患者门静脉期腹部CT直方图和小波特征的影响
- Author:
Gongbo XUE
1
;
Hongyan LIU
1
;
Guohua WANG
1
;
Zhen ZHANG
1
;
Xiao CHEN
1
;
Qiuyu DING
1
Author Information
1. 康复大学青岛医院 青岛市市立医院放射科，青岛 266011
- Publication Type:Journal Article
- Keywords:
Tomography, X-ray computed;
Texture analysis;
Abdomen;
Deep learning;
Image reconstruction algorithm
- From:
Chinese Journal of Radiology
2025;59(1):50-56
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the impact of deep learning image reconstruction (DLIR), adaptive statistical iterative reconstruction-veo (ASiR-V) and filtered back projection (FBP) on the histogram and wavelet features of portal venous phase abdominal CT in patients with hepatic tumor.Methods:The CT data of 68 patients with hepatic tumor who underwent enhanced CT scans were retrospectively collected. FBP, 30%ASiR-V, DLIR-L, DLIR-M and DLIR-H images were reconstructed. The images of portal venous phase were reconstructed with five algorithms, including FBP, ASIR-V at a level of 30% (ASiR-V 30%), DLIR at low (DLIR-L), medium (DLIR-M), and high (DLIR-H). Histogram and wavelet features were extracted from hepatic lesion, liver, spleen, kidney and erector spinae muscle, and compared using one-way analysis of variance or Kruskal-Wallis test. Two radiologists delineated the three-dimensional lesions independently and one of them repeated the delineation after one month. Intra-class correlation coefficients ( ICC) among five sets of images were calculated to evaluate the consistency of radiomics features of hepatic lesion. P<0.05 was considered to indicate statistical significance. Results:Most histogram and wavelet features extracted from hepatic lesion, liver, spleen, kidney and erector spinae muscle showed significant differences among five groups (all P<0.05). The number of features without significant differences decreased with the intensity of DLIR reconstruction increased. For histogram features, there were no significant differences of energy, mean, median, and total energy among five sets of images ( P>0.05). For wavelet features, there were no significant differences of mean and median among five sets of images ( P>0.05). The consistency of all histogram features was high except for the mean value of wavelet feature. The intra-and inter-observer ICC ranged from 0.756 to 1 and 0.767 to 1, respectively. Conclusion:Both 30%ASiR-V and DLIR at three levels algorithms had influence on the histogram and wavelet features of abdominal organs and hepatic tumors extracted from CT images in portal venous phase, and the effects expanded with the strengthening of levels. Median can be a reliable quantitative parameter for CT texture analysis of hepatic tumor.
