Feasibility of deep learning reconstruction algorithm combined with adual-low protocol for thoracoabdominal aortic CT angiography
10.3760/cma.j.cn112149-20241031-00653
- VernacularTitle:深度学习图像重建算法联合“双低”方案对胸腹主动脉CT血管成像的可行性研究
- Author:
Yingying HU
1
;
Yunpeng GAO
1
;
Yan CHEN
1
;
Nanxue LIANG
1
;
Yue LIN
1
;
Tongxi LIU
1
;
Peiyao ZHANG
1
;
Hongliang SUN
1
Author Information
1. 中日友好医院放射诊断科，北京 100029
- Publication Type:Journal Article
- Keywords:
Tomography, X-ray computed;
Angiography;
Deep learning;
Radiation dose;
Contrast agent
- From:
Chinese Journal of Radiology
2025;59(10):1149-1154
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the feasibility of deep learning reconstruction (DLR) algorithm combined with a dual-low protocol (low radiation dose and low contrast medium dose) for thoracoabdominal aortic CT angiography (CTA).Methods:This cross-sectional study prospectively enrolled 56 patients suspected of aortic diseases who underwent aortic CTA at China-Japan Friendship Hospital from June 2023 to June 2024. All patients were randomly divided into two groups: Group A (28 cases) underwent CTA with a tube voltage of 100 kVp, automatic tube current modulation (noise index=10), and a contrast agent dose of 80 ml (flow rate 5 ml/s), with images reconstructed using the three-dimensional adaptive iterative dose reduction algorithm (AIDR). Group B (28 cases) underwent CTA with a tube voltage of 80 kVp, automatic tube current modulation (noise index=25), and a contrast agent dose of 40 ml (flow rate 3.5 ml/s), with images reconstructed using either the deep learning reconstruction algorithm-Advanced intelligent Clear-IQ Engine (AiCE subgroup) or the AIDR (AIDR subgroup). Two physicians evaluated the image quality of the three groups subjectively and objectively. Objective evaluation metrics included CT values, image noise (SD), signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) at the ascending aorta, carina-level descending aorta, celiac trunk-origin abdominal aorta, and common iliac bifurcation abdominal aorta carina. Subjective evaluation metrics included image quality and noise scores. Comparisons among the three datasets (Group A, AiCE subgroup, AIDR subgroup) were performed using one-way ANOVA or the Kruskal-Wallis test, with appropriate post-hoc tests for pairwise comparisons.Results:No significant differences were observed in CT values of the ascending aorta, descending aorta, and abdominal aorta between Group A and the AiCE subgroup or the AIDR subgroup ( P0.05). However, significant overall differences were found in SD, SNR, and CNR values for the ascending aorta, descending aorta, and abdominal aorta ( P0.05). Pairwise comparisons revealed that, except for no significant differences in SD, SNR, and CNR values of the ascending and descending aorta between Group A and the AiCE subgroup, and no significant difference in SNR values of the ascending and abdominal aorta between Group A and the AIDR subgroup ( P0.05), all other intergroup comparisons showed statistically significant differences ( P0.05). Significant overall differences were also observed in image quality and noise scores between Group A and the AiCE and AIDR subgroups ( P0.05). Except for no significant differences in image quality and noise scores between Group A and the AiCE subgroup ( P0.05), all other pairwise comparisons showed statistically significant differences ( P0.05). Conclusions:The application of deep learning reconstruction algorithm combined with a dual-low protocol in thoracoabdominal aortic CTA can reduce radiation dose and contrast agent dose while maintaining diagnostic image quality, demonstrating significant clinical value for widespread adoption.
