|
KMID : 1149020210230010057
|
|
Journal of Korean Society of Computed Tomographic Technology
2021 Volume.23 No. 1 p.57 ~ p.65
|
|
|
A Study on Image Quality and Dose Comparison of Abdominal CT with Deep Learning Iterative Reconstruction Method and Model-Based Iterative Reconstruction Method
|
|
Shim Hyeon-Bo
Lee Kwang-Hyun
Kim Ryeon-Hee
Park Soon-Kyoo
Shim Ji-Na
|
|
|
Abstract
|
|
|
|
Deep Learning Iterative Reconstruction (DLIR), which was trained with an artificial neural network, was developed to secure the shortcomings of the traditional iterative reconstruction method and obtain a good quality image at a lower dose. The purpose of this study was to compare the effects of dose reduction and image quality evaluation using deep learning iterative reconstruction (True Fidelity; TF) and Siemens' Advanced Modeled Iterative Reconstruction (ADMIRE). Phantom study is a axial image with TF (TF-L, TF-M, TH-H) and ADMIR E (1,2,3,4,5) with CTDIvol (mGy) set equal to 9.5. Abdomen (A), Bone (B) R OI was set, and the patient study was a axial image applied with TF (TF-M, TF-H) and ADMIRE 2 in the abdominal contrast CT scan of 30 patients who performed the examination in our hospital. Abdomen aorta (A), Hepatic parenchyma (B), muscle (C), background (D) R OI was set to measure HU, SD values, and SNR , CNR were compared. In the Phantom study, TF-M was compared with ADMIRE 1, 3, 5. Noise was 221%, 139%, 40% lower in ROI A, 104%, 66%, 19% lower in B.(p<0.05) SNR was 70%, 59%, 33% higher in ROI A, 53, 42%, 19% higher in B.(p<0.05) In patient study, TF-M was compared with ADMIRE 2. Noise was 7% lower in ROI A but it was not statistically significant.(p=0.28) B was 21%, C was 45% lower.(p<0.05) SNR was the same in ROI A but not statistically significant.(p=0.70) B was 20%, C was 28% higher.(p<0.05) CNR was ROI A 22%, B 25%, C 22% higher.(p<0.05) In dose assessment, the average CTDIvol (mGy) of patients with TF applied was 4.73 ± 1.28, DLP (mGy cm) was 281.43 ± 79.22, and the average CTDIvol (mGy) of patients with ADMIRE was 6.06 ± 1.22, DLP (mGy cm) was 343.3 ± 81.34.(p<0.05) It can be seen that TF was lower in CTDIvol (mGy) by 28% and DLP (mGy cm) by 21% than in patients with ADMIRE.(p<0.05) In conclusion, TF images on abdominal CT showed no change in HU values than ADMIRE images, but Noise decreased and SNR and CNR increased. If the dose is the same, the TF image is applied has better image quality than the ADMIRE image. It can be said that it contributes significantly to reducing the patient's exposure dose.
|
|
|
KEYWORD
|
|
|
artificial neural network, deep learning iterative reconstruction method, model-based iterative reconstruction method, radiation dose, SNR, CNR
|
|
|
FullTexts / Linksout information
|
|
|
|
|
|
Listed journal information
|
|
|
|
|