MO-AB-BRA-06: Noise Index for Model Based Iterative Reconstruction (MBIR)? A Prospective Trial with 110 Human Subjects

Authors


Abstract

Purpose:

Noise index (NI) enables the mAs to be automatically determined to achieve the desired CT noise level. The scientific foundation behind the use of NI is the linear relationship between noise variance (σ2) and the 1/mAs. This relationship has been severely violated by the use of the highly nonlinear MBIR algorithm, making the traditional NI system no longer valid. This work studied how the quantitative relationship between σ2 and mAs should be modified in MBIR so that a new NI-mAs correspondence can be established.

Methods:

A comprehensive investigation using an anthropomorphic abdominal phantom, in-vivo swine, and an IRB-approved prospective human subject trial with a cohort of 110 subjects was performed. All scans were performed using a 64-slice CT scanner (Discovery CT750 HD, GE Healthcare) with typical abdominal protocols. Several mAs levels were used for the phantom and swine (10–350 mAs), while for human subjects two mAs levels were used. Both FBP and MBIR (Veo, GE Healthcare) reconstructions were performed. Values of σ2 were measured in relatively uniform regions such as liver and fat. The measured variances were related to mAs through the power-law relationship σ2 = α(mAs)^(β).

Results:

For liver, β is 1.0±0.10 for FBP and 0.4±0.12 for MBIR. The difference in β between the two methods is statistically significant (p<0.001). For patients with a given size (effective diameter∼28 cm), the α value is 1.2×10⁵ for FBP and 600 for MBIR. The R2 value fitting was 0.96±0.03.

Conclusion:

As long as the noise magnitude of a standard dose FBP image is known, in addition to predicting the noise magnitude of the corresponding MBIR image at the same dose, one can also predict the noise magnitude of the MBIR image at reduced dose. Therefore, it is both plausible and necessary to establish a new NI-mAs correspondence system for the MBIR.

This work was partially supported by an NIH grant R01CA169331 and GE Healthcare. K. Li, D. Gomez-Cardona, M. G. Lubner: Nothing to disclose. P. J. Pickhardt: Co-founder, VirtuoCTC, LLC Stockholder, Cellectar Biosciences, Inc. G.- H. Chen: Research funded, GE Healthcare; Research funded, Siemens AX.

Ancillary