288 Evaluation of iterative reconstruction algorithms in CT by means of low contrast detectability: A feasibility study

Abstract

Purpose With non-linear iterative reconstruction algorithms (IRA) conventional Fourier metrics are not applicable. Newly proposed metrics are complicated. This study aims to test the usefulness of straightforward low contrast detectability (LCD) evaluations based on a statistical method (1) for a task specific comparison of two IRA: raw data-based +knowledge based SAFIRE and knowledge-based (both Siemens, Erlangen). Methods Two recent CT systems of the same manufacturer with different IRA and different detector technology have been studied (Table 1). A homogeneous phantom (Siemens calibration phantom) has been acquired at 120 kV with exposure values ranging from 0.2 to 1.5 times the standard dose. Reconstruction has been carried out with three different clinical algorithms and slice thicknesses: smooth J30 (3.0 mm), medium J40 (1.5 mm), sharp J70 (1.5 mm). Based on a statistical approach a contrast-detail curve for 10 subsequent reconstructed slices has been determined, the mean curve calculated and interpolated. To summarize the data an overall index Image Quality Figure (IQF) has been defined. Results In both systems LCD improves steadly with CTDIvol. For filter J30 there is no difference in LCD between the two systems at clinical doses. At very low doses system 2 with ADMIRE scores slightly better. For filter J40 system 2 performs better at all dose levels, with an average dose reduction of −17%, which includes detector and IRA advantage. Conclusions The LCD method with adoption of the overall index IQF revealed sensitive to subtle variations in the image texture produced by different IRA. It has demonstrated to be a reliable tool for simple image quality assessment in CT quality control tasks. The data confirm that the novel algorithm ADMIRE does not bring a generalized image quality improvement or dose saving potential over SAFIRE, but that it acts selectively on the image quality, depending on the noise content of the region. In decreasing order of importance the possible gains depend on image kernel, IRA strength, IRA type. Download : Download high-res image (230KB) Download : Download full-size image Download : Download high-res image (295KB) Download : Download full-size image