Program Information
Understanding CT Iterative Reconstruction Kernels, Noise Characteristics and the Differences
G Ge1*, J Zhang2 , (1) VA Lexington Hospital, Lexington, KY, (2) University of Kentucky, Lexington, KY
Presentations
TU-RPM-GePD-I-6 (Tuesday, August 1, 2017) 3:45 PM - 4:15 PM Room: Imaging ePoster Lounge
Purpose: To better understand noise characterization of computed tomography (CT) images reconstructed with iterative reconstruction kernels and demonstrate their differences.
Methods: Unsubtracted and subtracted noise power spectra (NPS) and modulation transfer function (MTF) were measured using CT images acquired from a routine water phantom on Siemens Definition Edge and Force, respectively. Series of cutting-edge iterative reconstruction kernels in Sinogram Affirmed Iterative Reconstruction (SAFIRE I31f-I70f) and Advanced Modeled Iterative Reconstruction (ADMIRE I31f-I70f, and Br20s-Br70s in dual-source) were evaluated and compared at different strength settings (level 1 to 5).
Results: Each NPS demonstrates the expected shape of traditional back-projection filters. The peak frequency gradually shifts to lower frequency with the increased strength (from 1 to 5), corresponding to stronger smoothing effect. Low-frequency structured noise is observed in unsubtracted NPS for all kernels and is successfully removed in subtracted NPS. With the increases in filter sharpness, the NPS shifts to higher frequencies, preserving higher-frequency noise. The series of I kernels in SAFIRE and ADMIRE demonstrate comparable subtracted and unsubtracted NPS. However, the Br kernels in ADMIRE dual-source shows differences from I kernels. The NPS of a higher strength Br kernel (e.g, BR40s level 5) is closer to that of adjacent lower I kernels (e.g., I40f level 4), indicating that the same sharpness can be achieved with the same noise at a lower dose in Siemens Force. It is worth noting that kernels of the Siemens Definition family cannot be exactly mapped to kernels of Siemens Force. MTF shows spatial frequency preservation increases with filter sharpness. Over-enhancing is observed in both I and Br sharp kernels.
Conclusion: Analysis of NPS helps characterize image noise associated with various reconstruction kernels and identify their differences. A better understanding of what changes when new reconstruction algorithm is installed will facilitate their clinical applications.
Contact Email: