Application value of artificial intelligence iterative reconstruction algorithm in low-dose chest computed tomography
10.13491/j.issn.1004-714X.2025.06.017
- VernacularTitle:深度学习全模型迭代算法在胸部低剂量CT检查中的应用价值
- Author:
Xinyu LI
1
;
Mengxue LI
1
;
Shengnan FAN
1
;
Jingguo ZHANG
1
;
Jianxin GUO
2
;
Jun DENG
1
Author Information
1. National institute for radiological protection, Chinese center for disease control and prevention, Beijing 100088, China.
2. The First Affiliated Hospital of Xi’an Jiaotong University, Xi’an 710061, China.
- Publication Type:OriginalArticles
- Keywords:
Artificial intelligence iterative reconstruction algorithm;
Low-dose computed tomography;
Chest computed tomography;
Radiation dose
- From:
Chinese Journal of Radiological Health
2025;34(6):889-895
- CountryChina
- Language:Chinese
-
Abstract:
Objective To investigate the impact of the artificial intelligence iterative reconstruction (AIIR) algorithm on image quality in chest computed tomography (CT) at different radiation doses, and assess its value in reducing radiation dose during chest CT examinations. Methods A simulated chest phantom was scanned with 12 groups of tube voltages and milliampere-seconds, and the radiation dose was recorded for each group. The images of each group were reconstructed using seven methods: AIIR with noise levels 1-5, KARL iterative reconstruction, and filtered back projection (FBP). The CT values and standard deviations of soft tissue, thoracic vertebrae, pulmonary nodules, and the mediastinum were measured, with standard deviation representing image noise. Subjective evaluation of image quality was performed. The Friedman test was used to compare CT values among the seven reconstruction groups, a linear mixed model was employed for statistical analysis of image noise, and the Friedman test was also used for comparing subjective evaluation scores. Results The reconstruction algorithm, tube voltage, milliampere-seconds, and their interactions all showed statistically significant effects on image noise for the four tissues (F = 2.041-391.283, P < 0.05). Among the reconstruction algorithms, noise reduction capability decreased in the following order: AIIR levels 1-5, KARL, and FBP. The interaction between the reconstruction algorithm and tube voltage or milliampere-seconds indicated that AIIR exhibited improved noise reduction efficacy under low tube voltage and low milliampere-second conditions (|t| = 1.892-8.245, P < 0.05). In terms of subjective evaluation of image quality, there was no statistically significant difference among AIIR levels 3-5 (|Z| ≤ 0.567, P > 0.05), and the score of AIIR level 3 was significantly higher than those of AIIR level 1, AIIR level 2, FBP, and KARL level 2 (|Z| = 3.449-5.906, P < 0.05). Conclusion The AIIR reconstruction algorithm significantly reduced image noise in chest CT examinations. For improving image quality while maintaining image realism, AIIR level 3 is recommended, which can reduce the radiation dose by more than 75%. Furthermore, AIIR showed superior performance in noise reduction under low tube voltage and low milliampere-second conditions, demonstrating significant potential for reducing radiation dose.
