Feasibility study on diagnosis of pulmonary embolism using deep learning reconstruction algorithm in ultra-low radiation dose CT pulmonary angiography
10.3760/cma.j.cn112149-20240902-00533
- VernacularTitle:超低辐射剂量CT肺动脉成像联合深度学习重建算法诊断肺栓塞的可行性研究
- Author:
Jinjuan LU
1
;
Leilei SHEN
;
Zhenghong BI
;
Chun ZHOU
;
Yijing GUO
;
Weijian XU
;
Xiaodan YE
;
Mengsu ZENG
;
Mingliang WANG
Author Information
1. 上海市老年医学中心（复旦大学附属中山医院闵行院区）放射科，上海 201100
- Publication Type:Journal Article
- Keywords:
Pulmonary embolism;
Deep learning reconstruction;
CT angiography;
Radiation dose;
Image quality
- From:
Chinese Journal of Radiology
2025;59(8):886-893
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the feasibility of ultra-low dose (ULD) CT pulmonary angiography (CTPA) combined with deep learning reconstruction (DLR) in the diagnosis of pulmonary embolism (PE).Methods:This cross-sectional study prospectively enrolled 100 patients with suspected PE who underwent CTPA examination in Zhongshan Hospital Fudan University, and Shanghai Geriatric Medical Center from April to July 2024, and were randomly divided into the routine dose (RD) group and ULD group according to block randomization. Effective dose (ED) were calculated. The noise index of RD group and ULD group was set to 10 and 20, respectively. Other scanning parameters and contrast agent injection protocol were the same. The CT images of RD group were reconstructed using hybrid iterative reconstruction (HIR), while ULD images were reconstructed with HIR and DLR (ULD-HIR subgroup and ULD-DLR subgroup). The image quality of the three groups of images was subjectively evaluated (overall image noise, pulmonary artery display) and objectively evaluated [signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR) of the images] respectively. Finally, the diagnostic results of PE by the expert committee composed of three chief physicians were taken as the gold standard, and one physician with lower qualifications independently evaluated the diagnostic rate of PE in the three groups of images. Objective image quality parameters across the three groups were compared using ANOVA, with LSD post hoc test was used for multiple comparisons. Subjective scores among the three groups were analyzed using the Kruskal-Wallis H test, with Bonferroni corrected pairwise post hoc test was applied for multiple pairwise comparisons. Results:The ED in the RD group and ULD group were (2.7±0.5) mSv and (0.7±0.2) mSv, respectively, and the differences were statistically significant ( t=26.42, P<0.001). The overall differences in CT values of pulmonary arteries at all levels in the images of the RD group, the ULD-HIR subgroup, and the ULD-DLR subgroup were not statistically significant ( P>0.05).The RD group, ULD-HIR subgroup and ULD-DLR subgroup overall differences in SNR and CNR at all levels pulmonary arteries were statistically significant ( P<0.001), in which except for the differences in CNR and SNR values of the left pulmonary arterial trunk in the RD group and the ULD-HIR subgroup, and SNR values of basal segment pulmonary artery of the lower lobe of the left lung, which were not statistically significant ( P>0.05), the differences of the rest of the indexes in the pairwise comparisons between the groups were statistically significant ( P<0.05). The overall differences in the subjective scores of image pulmonary vascular display and image noise in the RD group, ULD-HIR subgroup and ULD-DLR subgroup were statistically significant ( P<0.001), except that the differences in the subjective scores of image pulmonary vascular display in the ULD-DLR subgroup were not statistically significant when compared with that of the RD group ( P>0.05) and that of the rest of the metrics in the between-groups two-by-two comparisons were all statistically significant ( P<0.05). The difference in diagnostic rates of PE in the pulmonary artery trunk, lobe and segmental levels in the images of the RD group, ULD-HIR subgroup and ULD-DLR subgroup was not statistically significant ( P>0.05). Conclusions:DLR can significantly reduce the radiation dose of CTPA examination. Even at ultra-low radiation dose, its image quality is still better than HIR reconstruction at conventional doses and preserve diagnostic accuracy of PE at the lobe level and segment level.
