High contrast resolution(HCR) is an important performance parameter of the computed tomography(CT) scanner. The assessment of HCR is regularly performed by measuring the modulation transfer function (MTF). Generally, MTF is calculated from point spread function (PSF) which can be approximated by the point source response in CT phantom images. In the traditional method of calculating MTF, a rectangle region of interest (ROI) is usually placed on the scanned image to crop the point source response. However, the sampling interval of phantom images obtained in practice is often coarse (about 0.40 to 0.50 mm), which means only a few pixels lie in the vicinity of the point source. Therefore, the size of ROI affects the accuracy of MTF when image noise exists. In this study, based on the coarsely sampled CT images of Catphan500 phantom, we analyze how the ROI size influenced the accuracy of MTF. Then the traditional method was improved by optimizing the ROI size. We found that when the ROI size was close to the size of point source response, the obtained MTF was more accurate. We verified the validity of improved method with Catphan500 images from five different medical CT scanners. The results showed that the cut-off frequency of obtained MTF had a good consistency with cut-off frequency evaluated by experts. The improved method is particularly useful for routine HCR assessment of the CT scanner.