Automated objective determination of modular transfer function (MTF) for quality control of CT scanners

Abstract

Problem: The Modular Transfer Function (MTF) is an established objective metric to characterise the spatial resolution of Computer Tomography (CT) systems. Both free and commercial tools exist to perform the calculation, however a disadvantage of the latter is often a lack of transparency in the exact algorithmic steps performed. Another disadvantage of both categories of tools is that they are often designed to be applied manually on single axial images, making them difficult to integrate into semi-automated, quality control platforms for data collection, analysis and reporting. This work sought to develop an in-house tool that both provides transparency and able to automate bulk images. Method: The tool was developed in Python to calculate the MTF based on cross-sectional images using the wire insert in CTP591. Five images were collected from a Siemens Somatom Edge Plus (120 kVp, 250 mAs, 5 mm slice thickness) and reconstructed using filtered backprojection with convolution kernels Br32s, Hr32s, Hr38s, Hr45s, Hr60s. An important parameter choice in MTF calculation was the size of the ROI around the wire, which was investigated. Finally, MTF critical frequencies at 50%, 10%, and 2% were compared to a commercial tool, CT AutoQA Lite. Results: The MTF profile changed as a function of reconstruction kernel. Although this was expected, at a very high level of sharpness (Hr60s), the MTF obtained was highly affected by the noise in the image. The results of the spatial frequency from the in-house developed tool, although were not exactly the same as the results obtained from CT AutoQA Lite followed the same trend. The plots obtained, were of the same profile. This tool was deemed best for automation as unlike CT AutoQA Lite it was designed to work on individual modules rather than requiring the full scan of the phantom. The tool was designed to provide the data points, which can be integrated into an automated platform which could re-plot the data points as part of an auto generated report. Conclusion: An in-house tool to calculated MTF was developed. It was understood that there are many approaches to calculate MTF and that these will have an effect on the profile. Further work is required to fully develop the in-house tool. There should be an appreciation that different tools may result in different MTF critical frequencies.