An effective method to verify line and point spread functions measured in computed tomography

This study describes an effective method for verifying line spread function (LSF) and point spread function (PSF) measured in computed tomography (CT). The CT image of an assumed object function is known to be calculable using LSF or PSF based on a model for the spatial resolution in a linear imaging system. Therefore, the validities of LSF and PSF would be confirmed by comparing the computed images with the images obtained by scanning phantoms corresponding to the object function. Differences between computed and measured images will depend on the accuracy of the LSF and PSF used in the calculations. First, we measured LSF in our scanner, and derived the two-dimensional PSF in the scan plane from the LSE Second, we scanned the phantom including uniform cylindrical objects parallel to the long axis of a patient's body (z direction). Measured images of such a phantom were characterized according to the spatial resolution in the scan plane, and did not depend on the spatial resolution in the z direction. Third, images were calculated by two-dimensionally convolving the true object as a function of space with the PSF. As a result of comparing computed images with measured ones, good agreement was found and was demonstrated by image subtraction. As a criterion for evaluating quantitatively the overall differences of images, we defined the normalized standard deviation (SD) in the differences between computed and measured images. These normalized SDs were less than 5.0% (ranging from 1.3% to 4.8%) for three types of image reconstruction kernels and for various diameters of cylindrical objects, indicating the high accuracy of PSF and LSF that resulted in successful measurements. Further, we also obtained another LSF utilizing an inappropriate manner, and calculated the images as above. This time, the computed images did not agree with the measured ones. The normalized SDs were 6.0% or more (ranging from 6.0% to 13.8%), indicating the inaccuracy of the PSF and LSE We could verify LSFs and PSFs for three types of reconstruction kernels, and demonstrated differences between modulation transfer functions (MTFs) derived from validated LSFs and inaccurate LSFs. Our technique requires a simple phantom that is suitable for clinical scanning, and does not require a particular phantom containing some metals or specific fine structures, as required in methods previously used for measurements of spatial resolution. Therefore, the scanned image of the phantom will be reliable and of good quality, and this is used directly as a confident reference image for the verification. When one obtains LSF, PSF or MTF values, verification using our method is recommended. Further, when another method for the measurement of LSF and PSF is developed, it could be validated using our technique, as illustrated in the method proposed by Boone [Med. Phys. 28, 356-360 (2001)] and used in this paper.