Deep learning image reconstruction algorithm combined with a large reconstruction matrix for low-dose CT screening of lung nodules
10.3969/j.issn.1002-1671.2025.11.028
- VernacularTitle:深度学习图像重建算法联合大重建矩阵行肺结节低剂量CT筛查
- Author:
Changyu DU
1
;
Wei WEI
1
;
Mengting HU
1
;
Jingyi ZHANG
1
;
Qiye CHENG
1
;
Jian HE
1
;
Anliang CHEN
1
;
Yijun LIU
1
Author Information
1. 大连医科大学附属第一医院放射科,辽宁 大连 116011
- Publication Type:Journal Article
- Keywords:
computed tomography;
lung nodule;
deep learn-ing image reconstruction;
reconstruction matrix;
image quality
- From:
Journal of Practical Radiology
2025;41(11):1886-1890
- CountryChina
- Language:Chinese
-
Abstract:
Objective To explore the application value of deep learning image reconstruction(DLIR)algorithm combined with a large reconstruction matrix in lung nodules screening using low-dose computed tomography(LDCT)of the chest.Methods Patients who underwent LDCT scans were prospectively enrolled.The control group(group A)used the iterative reconstruction(IR)algorithm(Karl)with a reconstruction level of Karl 5,reconstructed images of 512×512(group A1)matrix,and 1 024 × 1 024(group A2)matrix.The experimental group employed DLIR combined with 512×512(group B)matrix and 1 024 × 1 024(group C)matrix for image reconstruction at levels 1-5,which were recorded as groups B1-5 and groups C1-5.The CT values and standard deviation(SD)values of the lung parenchyma and tracheal air were measured,and the signal-to-noise ratio(SNR)was calculated.The overall lung image quality was scored on a Likert 5-point scale,and the subgroup with the best lung image quality was selected.The lung nodule detec-tion rate and clarity were compared with group A1.Results Under the same reconstruction matrix,the CT values of the tracheal air and lung parenchyma in the experimental group showed no significant difference compared to the control group,while the SD values were lower and SNR were higher(P＜0.05).Within groups B and C,as the DLIR level increased,the SD values of the tracheal air and lung paren-chyma gradually decreased,and SNR gradually improved(P＜0.05).Subjective scores for the image quality in groups B and C initially increased and then decreased,with group B3 and group C4 showed the best image quality.No difference was observed in objective eval-uation between the two groups,but the subjective image quality score of group C4 was superior to group B3(P＜0.05).Subjective eval-uation of lung nodule display in group C4 was better than in group A1(P＜0.05).Conclusion DLIR algorithm combined with a large reconstruction matrix is feasible for lung nodules screening in chest LDCT,reducing image noise while improving lung nodules clarity,demonstrating significant clinical value.
