Admire vs. Safire: Objective comparison of CT reconstruction algorithms and their noise properties

TL;DR

This study objectively compares the noise reduction and texture preservation of Siemens' CT reconstruction algorithms SAFIRE and ADMIRE, revealing that ADMIRE more effectively reduces noise near edges but does not significantly improve noise structure overall.

Contribution

It provides an objective, quantitative comparison of SAFIRE and ADMIRE algorithms' noise properties using NPS and standard deviation maps, highlighting differences in noise removal efficiency.

Findings

ADMIRE removes noise near edges more effectively than SAFIRE.

No significant difference in noise structure preservation between ADMIRE and SAFIRE.

At low dose, both algorithms produce more blotchy textures, with ADMIRE showing a more prominent shift in NPS.

Abstract

Purpose: Siemens has developed several iterative reconstruction (IR) algorithms on their CT scanners. SAFIRE is available on most of their CT scanners. The latest algorithm, ADMIRE, is available on their newest high-end CT scanners. The aim of our study was to compare the noise reduction properties of the two IR algorithms using objective methods. Methods and Materials: The homogeneous module of the Catphan phantom was scanned on a Siemens AS+ and a Siemens Flash CT scanner using an axial abdomen protocol with fixed tube current at two dose levels. The images were reconstructed with an abdomen filter (B30) using filtered back projection (FBP) and a low, medium, and high level of SAFIRE or ADMIRE. Noise Power Spectrum (NPS) curves were calculated using these images. Then, an anthropomorphic abdomen phantom (Kyoto Kagaku PH-5) was scanned using the same setup and exposure parameters. Fifty axial images at the same slice location were used to calculate inter-image standard deviation maps. Results: At full dose, the median values of the NPS curves were similar for both scanners at all IR levels. At low dose the median values of the NPS curves were generally shifted towards lower spatial frequencies, usually resulting in a more blotchy image texture. This shift was more prominent for ADMIRE compared to SAFIRE for all IR levels. Based on the inter-image standard deviation maps of the anthropomorphic phantom, ADMIRE removed noise near edges more efficiently than SAFIRE. Conclusion: No significant improvement in maintaining noise structure were found for the ADMIRE algorithm. Based on the inter-image standard deviation maps, ADMIRE removed noise near edges more efficiently than SAFIRE.