Abstract
PURPOSE: This study aimed to develop and validate a method for characterizing the spatial resolution of clinical chest computed tomography (CT) sequence images. METHODS: An algorithm for characterizing spatial resolution based on clinical chest CT sequence images was developed in Matlab (2021b). The algorithm was validated using CT sequence images from a custom-made chest automatic tube current modulation (ATCM) phantom and clinically reconstructed chest CT sequence images. A region of interest (ROI) was automatically established at the edges of CT image subject to calculate the edge spread function (ESF). The ESF curves from consecutive CT images within the same sequence were fitted into a curve, and the line spread function (LSF) was derived through differentiation. A Fourier transformation of the LSF curve was conducted to obtain the modulation transfer function (MTF). The method's effectiveness was verified by comparing the 50% MTF and 10% MTF values with those calculated using IndoQCT (22a) software. The method was also applied to clinical CT images to calculate MTF values for various reconstructions, confirming its sensitivity by determining spatial resolution of clinically reconstructed images. RESULTS: Validation experiments based on the phantom CT sequence images demonstrated that the MTF values calculated using the proposed method had an average difference of within ± 5% compared to the results obtained with IndoQCT. Validation experiments with clinical CT sequence images indicated that the method effectively reflects differences and variations in spatial resolution of images under different reconstruction kernels, with the MTF values for B10f-B50f and D10f-D50f exhibiting a consistent increase. CONCLUSION: A method for measuring spatial resolution using clinical chest CT sequence images was developed. This method provides a direct means of spatial resolution characterization for clinical CT datasets and a more accurate representation of CT imaging quality, effectively reflects variations across different reconstruction convolution kernels, demonstrating its sensitivity.