The effects of deep-learning image reconstruction algorithm on image quality of lower extremity CT angiography with low kV and reverse flow direction scanning
10.3760/cma.j.cn112149-20220225-00159
- VernacularTitle:深度学习重建算法对低kV逆血流扫描下肢动脉CT血管成像图像质量的影响
- Author:
Yilin CHEN
1
;
Yuanfen LIU
;
Lili WANG
;
Xiongxin YE
;
Yunjing XUE
Author Information
1. 福建医科大学附属协和医院放射科，福州 350001
- Keywords:
Tomography, X-ray computed;
Artificial intelligence;
Deep learning;
Lower extremity artery;
Image quality
- From:
Chinese Journal of Radiology
2022;56(11):1188-1194
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate impacts of a deep learning image reconstruction (DLIR) algorithm on image quality of lower extremity CTA with low kVp and reverse flow direction scanning.Methods:From January 2021 to March 2021, fifty patients with suspected lower extremities diseases and received lower extremity CTA with low kVp and reverse flow direction scanning in Union Hospital affiliated to Fujian Medical University were retrospectively collected in this study. Six groups of CT images were reconstructed at the thickness of 0.625 mm using two algorithms including ASIR-V of three blending ratios (ASIR-V 20%, ASIR-V 50% and ASIR-V 80%) and DLIR of three strengths (DLIR-H, DLIR-M and DLIR-L). Regions of interest (ROIs) were placed on proximal abdominal aorta (AA), distal AA, left and right common iliac arteries, left and right femoral arteries (upper segment), left and right superficial femoral arteries (middle segment), left and right popliteal arteries. The CT value and SD value were measured for each group; the signal-noise ratio (SNR) and contrast-noise ratio (CNR) were calculated. The lower extremity CTA was divided into four segments, and the subjective evaluation was independently performed on noise and sharpness using 4 points scales by two radiologists. One-way analysis of variance was utilized to evaluate the differences in subjective scoring and objective parameters among the six groups.Results:For all arteries segments, with the increase of blending ratios for ASIR-V and reconstruction strength of DLIR, the SD values were reduced while SNR and CNR were increased (all P<0.05). Among the six groups, DLIR-H and ASIR-V80% images had lowest SD as well as highest SNR and CNR (all P<0.05). In comparison to ASIR-V20% and ASIR-V50% images, DLIR-H images showed lower SD, higher SNR and CNR values (all P<0.05). There were no statistical differences between ASIR-V80% and DLIR-H images in SD, SNR and CNR values (all P>0.05). Subjective scoring results showed that the DLIR-H images displayed the best noise performance for the entire lower extremity arteries from AA to the foot artery, and the sharpness scores of DLIR-H images were also significantly higher than ASIR-V80% (all P<0.05). Conclusion:DLIR can significantly reduce the image noise and improve the image quality in CTA for abdominal aorta to lower extremity arteries. DLIR-H showed the greatest noise reduction ability and the best effect balancing noise and sharpness, providing highest image quality.
