Investigation of patch-based modulation transfer function (MTF) prediction framework in radiography

Abstract The modulation transfer function (MTF) is a well-established in imaging systems and has played a significant role in assessing the performance of information transfer in accordance with the spatial frequency. Exist methods for measuring MTF require a pinhole, or a slit camera, an edge phantom, etc. These depend on the imaging system and exposure conditions. To overcome these problems, we investigated the proposed patch-based MTF prediction framework using a deep-learning algorithm in radiography. This proposed method directly allows for the predicted MTF from a single image without the need for additional devices and work. We used 800 to 1500 projections for supervised learning and selected many patches, including the edge information from input image for predicting the MTF, for which we used the Kendall's rank correlation. The quantitative evaluation performed included an intensity profile and a root-mean square error. In addition, we implemented the nonblind deblurring process to verify its image performance using the predicted MTF. Our methods successfully predicted the MTF and restored them from the degraded image without such artifacts as the ringing artifact that can occur when using the incorrect blur kernel. Consequently, the simulation and experimental results indicate that the proposed framework is useful to predicting the accurate MTF and is effective in improving radiographic image quality.