The purpose of this study was to reveal the optimal function for regression of spectral Hounsfield Unit (HU) curves. The optimization procedure consists of the following steps: 1) obtaining dual energy CT (DECT) images of the RMI 467 phantom, 2) obtaining virtual monochromatic images from DECT images, 3) mapping each region of interest (ROI) to a phantom rod on virtual monochromatic images, 4) obtaining spectral HU curves for all rods, 5) regression of spectral HU curves using various functions, including linear, quadratic polynomial, cubic polynomial, quartic polynomial, quintic polynomial, sextic polynomial, septic polynomial, exponential, corrected exponential, bi-exponential, and logarithm, and 6) calculating the coefficients and the Akaike Information Criterion (AIC) of the functions listed above. Results indicated that the quintic polynomial function is suitable for analyzing the regression of spectral HU curves. The coefficients generated by the quartic or higher order polynomial functions were significantly higher than those generated by other functions (p<0.05). The median AIC of the quintic polynomial was the lowest among all functions. Therefore, we conclude that the quintic polynomial is the best function to use for the regression of spectral HU curves.
Keywords: Akaike Information Criterion (AIC); approximate function; dual-energy CT; optimization; spectral HU curves.