Room: Exhibi Hall | Forum 6
Purpose: The purpose of this study was to investigate the influence of computed tomography (CT) slice thickness and convolution kernel on the variability of texture features as a function of increasing CTDIvol on a cadaveric human liver.
Methods: A liver was embedded in 1% agar solution and underwent multiple scans on a single scanner with various acquisition parameters. Scans were split into four groups: â€œstandardâ€? and â€œdetailâ€? convolution kernels, and 1.25-mm and 5-mm slice thicknesses. Each group was evaluated at CTDIvol values of 5 and 13 mGy. Square 64x64-pixel regions of interest (ROIs) were placed in 7 and 17 consecutive slices in 5-mm and 1.25-mm reconstructions, respectively. ROIs were placed in the same anatomic area of the liver for each scan to evaluate first-order and gray-level co-occurrence matrix (GLCM) features. To quantify variability in the data across time, reference scans with identical settings were acquired immediately before and after the imaging session. The percent change in each feature between reference scans was the minimum threshold for meaningful differences in texture features with increasing CTDIvol. An F-test was applied to assess differences in variance between the 5-mGy and 13-mGy CTDIvol scans within each group.
Results: Among the 1.25-mm CT scans, both first-order and GLCM features had significantly lower variance at high CTDIvol, while the 5-mm group had no statistical difference in variance. The GLCM features inertia, sum variance, and difference variance had a statistical difference in the standard and detail convolution kernel groups. The average percent differences from 5-mGy to 13-mGy CTDIvol scans for first-order and GLCM features were -72.7% Â± 130% and -30.7% Â± 31%, respectively.
Conclusion: Many features, particularly GLCM features, have shown to vary with changing CTDIvol, and the degree of this variability has been shown to depend on slice thickness and convolution kernel.
Reconstruction, CT, Quantitative Imaging