OA139] Evaluation of two model based iterative reconstruction algorithms in computed tomography

Abstract

Purpose To investigate the mode of action of two commercial model based (MB) CT reconstruction algorithms by studying several standard Fourier and alternative metrics. Methods Three CT systems featuring two statistical and two MB reconstruction algorithms were studied: Brilliance Elite iCT 256 (iDose4/IMR; Philips, The Netherlands), Somatom AS64 (Safire) and Somatom Drive (Admire; Siemens, Erlangen). Three phantoms were used for acquisitions (Catphan, The Phantom Laboratory, USA; body-shaped commercial density phantom; cylindrical water-filled phantom) varying dose (CTDIvol) and kV. Images were reconstructed at 5 different thicknesses with Filtered Back Projection (FBP), statistical and MB iterative reconstruction algorithms of different type (soft, medium, sharp) and strength. Spatial resolution was evaluated with Modulation Transfer Function (MTF) and Target Transfer Function (TTF), noise with Standard Deviation. Noise behaviour was analysed with 3D and 2D Noise Power Spectrum. Furthermore, Blur and Low Contrast Detectability (LCD) were investigated. Image quality of MB reconstruction algorithms has been compared to the clinically corresponding FBP and statistical reconstruction algorithms. Results Both MB reconstruction algorithms exhibited a slight improvement for MTF, being the difference larger for IMR. For Admire the difference is kernel independent and increases slightly with iterative strength. TTF curves and Blur confirm this trend for high density objects end present a sharpness reduction for low density details. For Admire noise reduction is dose neutral, increases only slightly with kernel sharpness (difference Conclusions MB reconstruction algorithms produce an improvement in image quality and allow dose reduction. To optimize image quality each algorithm requires vendor specific reconstruction parameters. Although MB reconstruction algorithms are non-linear, LCD evaluation demonstrated partially meaningful. Further investigation regarding model observer applications and other metrics as the detectability index d’ should be carried out.