Study on deep learning reconstruction algorithm to improve image quality in low dose abdominal and pelvic CT angiography
10.3760/cma.j.cn112149-20231117-00397
- VernacularTitle:深度学习重建算法改善腹盆部血管低剂量扫描图像质量的研究
- Author:
Tingting QU
1
;
Le CAO
;
Yannan CHENG
;
Lihong CHEN
;
Yanan LI
;
Yinxia GUO
;
Jianying LI
;
Jian YANG
;
Jianxin GUO
Author Information
1. 西安交通大学第一附属医院医学影像科，西安 710061
- Keywords:
Tomography, X-ray computed;
Reconstruction algorithm;
Deep learning
- From:
Chinese Journal of Radiology
2024;58(6):647-652
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To investigate the practicality of TrueFidelity deep learning reconstruction algorithm in low-dose abdominal and pelvic CT angiography (CTA).Methods:The patients who required abdominal and pelvic CTA were prospectively included at the First Affiliated Hospital of Xi′an Jiaotong University from June 2020 to March 2021. All patients underwent low-dose CTA with a tube voltage of 80 kV and smart tube current modulation (100-720 mA). Images were reconstructed using the traditional FBP, adaptive statistical iterative reconstruction with a strength of 50% (ASIR-V 50%), TrueFidelity with medium (TF-M) and high (TF-H) strength. The CT value and standard deviation (SD value) of the abdominal aorta, psoas major muscle and subcutaneous fat in the same layer were measured, signal to noise ratio (SNR) and contrast to noise ratio (CNR) were calculated. We also introduced the measurement of skewness of CT value in psoas major muscle with uniform density. The above indexes of the four groups of reconstructed images were compared. A 5-point scoring method was used to evaluate the granularity, fuzziness and beam-hardening artifacts of all images. Objective measurement indicators, such as CT values, were tested by repeated measure ANOVA with the Bonferroni post hoc test.Results:There were forty-six patients in the study. The volume CT dose index of the scan was low at (1.09±0.31)mGy. There was no significant difference in CT values of vessels and muscles between the four groups ( P>0.05), but there was a significant difference in SD value( P<0.001). The SD value of the FBP group was the largest and that of the TF-H group was the smallest. The difference between SNR and CNR was statistically significant ( P<0.001), and the overall trend was opposite to that of the SD value. There was no significant difference in the skewness between the four groups. The granularity score of the FBP group was the largest, that of the TF-H group was the smallest, and there was a significant difference among the four groups. The score of fuzziness in the TF-H group was slightly higher than that in the other three groups, but there was no significant difference. The beam-hardening artifact score of FBP and ASIR-V 50% group was the worst, and the TF-H group was the best ( P<0.001). Conclusions:Compared with FBP and ASIR-V, TrueFidelity reconstruction algorithm provides better image quality (comprehensively considering image noise, fuzziness, uniformity, and hardening artifacts) in low-dose CT scanning of abdominal and pelvic vessels, and TF-H has the best image quality.
