Abstracts

RETURN TO ABSTRACT LISTING


E1626. Comparison of Noise Reduction Technologies Using Central Frequency Ratio and Noise Magnitude Ratio of Noise Power Spectrum
Authors
  1. Toshihiro Ishihara; National Cancer Center Japan
  2. Yuta Miyamae; National Cancer Center Japan
  3. Ikuya Ishii; National Cancer Center Japan
  4. Akira Hasegawa; AlgoMedica, Inc.
  5. Pan Tinsu; MD Anderson Cancer Center
  6. Miyuki Sone; National Cancer Center Japan
  7. Tomohiko Aso; National Cancer Center Japan
Background
There are a variety of noise reduction technologies (NRT) including iterative reconstructions (IR) and newer deep-learning based reconstructions (DLR) from vendors1. To access behaviors of these NRT, noise power spectrum (NPS) is a useful tool. The square of the area under the curve (AUC) of NPS is the noise magnitude and the curve shape is associated with noise texture. Many studies using NPS to compare NRTs just employed the visual comparison of multiple NPS curves2,3. Even in the assessment of one NRT, the NPS curves of data reconstructed with/without the NRT have to be relatively evaluated. When multiple different NRTs are to be compared, the comparison of NPSs becomes more complex and subjective if it is done visually. To make the NPS comparison of NRTs simple and objective, we have introduced central frequency ratio (CFR) and noise magnitude ratio (NMR)4, which are two metrics derived from ratios of NPS in central frequency and AUC before and after noise reduction, respectively. By connecting CFR-NMR plot of each strength of a NRT, a CFR-NMR curve of the NRT can be drawn. CFR-NMR curves helps to compare different noise reduction technologies without subjectivity. The purpose of this study is to compare a vendor’s popular noise reduction technologies using CFR-NMR curves.

Educational Goals / Teaching Points
1. To understand CFR and NMR of NPS 2. To understand how to draw CFR-NMR curve 3. To compare NRTs using CFR-NMR curves

Key Anatomic/Physiologic Issues and Imaging Findings/Techniques
The Catphan phantom was scanned with 120 kVp and CTDIvol 2.2mGy (Body16) by Canon Aquilion One. And, CT images were reconstructed with a filter back projection (FBP), AIDR3D, FIRST, and AiCE, and NPS was calculated5 from each reconstructed data set, and CFR-NMR curve was drawn.

Conclusion
Of the three, AIDR3D Strong and AiCE Body Mild are similar in terms of noise reduction and noise textures in both XY plane and Z direction. Compared to AIDR3D and AiCE, FIRST has more frequency shift in XY plane but less in Z direction.