Purpose
Image reconstruction technologies of computed tomography (CT) have been rapidly progressing.
Model-based iterative reconstruction (MBIR) is an image reconstruction technique that accurately reproduces CT values without filtered back projection. While MBIR can improve the image quality,it usually requires higher computational power and longer computational time [1].
Recently, advanced intelligence clear-IQ engine (AiCE) that using deep learning technology was developed.
The teaching data used for deep learning reconstruction (DLR) training is high-quality CT images reconstructed with MBIR whose parameters are adjusted to obtain the best image...
Methods and materials
To evaluate low-contrast detectability, we calculated low-contrast object specific contrast-to-noise ratio (CNRLO) from CT images. CNR is general to evaluate low contrast detectability but frequency characteristic is not taken into account. Then CNR is not able to use for evaluation of images using IR. CNRLOis considered frequency characteristic for evaluation low contrast detectability in image using IR [5].
Fig. 1shows theformula of CNRLOand axial slice of the human phantom depicting the regions of interest (ROI). ROIMand ROIBare CT values measured in the simulated nodule andliverparenchyma,...
Results
CTDIvolat normal dose was 18.7 mGy and at respective doses were 0.9, 4.9, 10.2 and 15.6 mGy.
Fig. 3 shows NPS resultsforeAIDR and AiCE at each dose. NPS obtained with both reconstructed images were improved as dose increased. Fig. 4 shows NPS at 18.7 mGy reconstructed with eAIDR and AiCE. As same radiation dose, although NPS obtained with both reconstruction methods were equivalent values at low-frequency range, NPS obtained with AiCE was dramatically improved at over middle-frequency range.
Fig. 5 shows CNRLOresults for eAIDR and...
Conclusion
NPS was improved at over middle frequency range, however it was not obtained differences between eAIDR and AiCE at low frequency range.
CNRLOwas not obtainedsignificant differences at each radiation dose between eAIDR and AiCE.
For these results, as AiCE could reduce noise of high frequency range, images with AiCE got better impression than eAIDR.AiCEwas not superior toeAIDRin terms of low-contrast detectability at the same radiation doses.
We should use AiCE carefully to detect low contrast objects in low dose abdominal CT images.
Personal information and conflict of interest
References
[1]Samei E, Richard S.et al.Assessment of the dose reduction potential of a model- based iterative reconstruction algorithm using a task-based performance metrology.Med Phys. 2015;42(1):314-323.
[2] Higaki T, Nakamura Y, et al.Deep Learning Reconstruction at CT: Phantom Study of the Image Characteristics. Academic Radiology 2020;27(1):82-87.
[3] Akagi M, Nakamura Y, et al. Deep learning reconstruction improves image quality of abdominal ultra-high-resolution CT
[4] Higaki T, Nishimaru E, et al. Radiation dose reduction in CT using Deep Learning based Reconstruction (DLR): A phantom study. ECR 2018;C-1656
[5]Urikura...