(Sunday, 7/29/2018) 2:05 PM - 3:00 PM
Room: Davidson Ballroom B
Purpose: To study radiomic features’ reliability on abdominal CT images reconstructed with a wide-range of image acquisition parameters.
Methods: The ACR CT phantom was scanned on a GE Discovery 750HD scanner using an adult abdomen protocol. Twenty-four different acquisition settings, combinations of four tube currents (25, 50, 100, 200 eff.mAs), three slice thicknesses (1.25, 2.5, 5 mm), and two reconstruction kernels (STANDARD and SOFT), were used. CT Images of two materials in the ACR phantom module 1 (Polyethylene (-95 HU) and acrylic (120 HU)) were selected to generate ROIs for feature calculations. We programmed a noise-free, computer-generated phantom image that reproduced these objects and the background contained in the ACR phantom. Feature values calculated from the computer-generated phantom image were used as reference values. Feature reliability was defined as the difference between calculated value from CT scan image and the reference value. Eight widely-used radiomic features, mean, standard deviation, skewness, kurtosis, GLCM(Gray Level Co-Occurrence Matrices)-energy, GLCM-contrast, GLCM-correlation, GLCM-homogeneity, were calculated. Linear regression model was employed to fit the calculated values from CT images and the referenced feature values. The feature reliability was quantified by the fitness (R-Square) and R-Square ≥ 0.85 was defined as high reliability.
Results: The reliability of the features of mean and standard deviation were high across all imaging parameters. At 200 eff.mAs, except GLCM-homogeneity, all features showed high reliability. Whereas at 25 eff.mAs, all features (except mean and standard deviation) had low reliability. In general, from highest to lowest, ranked the reliability of GLCM-energy, correlation, contrast and homogeneity across all imaging parameters.
Conclusion: CT imaging parameters could affect radiomic features differently. Higher reliability was observed for the features extracted from the images reconstructed at larger tube current and thicker slice settings. Standard and Soft reconstruction kernels showed similar results across varying tube currents and slice thicknesses.
Keywords
Feature Extraction, Quantitative Imaging, CT
Taxonomy
IM/TH- Image Analysis (Single modality or Multi-modality): Imaging biomarkers and radiomics
Contact Email
