Sphere phantom approach to measure MTF of computed tomography system using convolutional autoencoder network

Abstract

Image quality assessment is important task to maintain and improve the imaging system performance, and modulation transfer function (MTF) is widely used as a quantitative metric describing the spatial resolution of an image for fan beam computed tomography (CT) system. In MTF measurement, although fine wire and edge objects are generally used, there are difficulties in precise phantom alignment. To overcome this limitation, a sphere object was considered as an alternative phantom due to spherically symmetric property. However, the sphere phantom is more suitable for measuring 1D MTF along the particular direction than 2D MTF for fan beam CT system. In this work, we proposed a sphere phantom approach to measure whole 2D MTF of fan beam CT system using convolutional autoencoder network. We generated projection data of point and sphere objects, and reconstructed using filtered back-projection (FBP). The reconstructed point image was regarded as an ideal 2D point spread function (PSF). The ideal 2D modulation transfer function (MTF) was calculated by taking Fourier transform of the ideal 2D PSF. To measure 2D MTF, we divided the Fourier transform of reconstructed sphere phantom by that of ideal sphere object. The estimation errors caused by the inverse filtering were corrected using proposed convolutional autoencoder network. After applying the network, the estimated 2D MTF shows a good agreement with the ideal 2D MTF, indicating that the convolutional autoencoder network is effective for measuring 2D MTF of fan beam CT system.