Comparison of Iterative Reconstruction algorithms on PET-CT devices : Iterative CT -reconstruction

Abstract

Medical imaging utilizes radiation to examine a patient’s anatomy and internal structures. One imaging method is CT imaging, where an X-ray tube and detectors on the opposite side rotate around the patient to get images describing the patient’s internal structures in 3-D. The PET-CT device combines PET and CT imaging, where PET imaging provides information on the patient’s anatomy, and CT imaging can be used for attenuation correction of PET-imaging. In the imaging methods using ionizing radiation, the used dose affects the quality of the image. However, based on the purpose of the imaging, the dose should be as low as reasonably possible, according to the purpose of the image. For example, PET-CT imaging uses CT images for attenuation correction for the PET -image, so low-dose CT is enough. On the other hand, when the CT-image is needed to describe the patient’s anatomy to make a diagnosis, a higher dose is needed to distinguish the different structures well enough so that the patient can be diagnosed. The iterative reconstruction algorithms are mathematical methods in CT-device that improve the quality of CT-images and reduce the noise. Partly, through iterative reconstruction methods, the dose levels used in the scans can also be reduced.

This work compares two iterative reconstruction methods, ADMIRE and SAFIRE, which are algorithms in Siemens PET-CT-devices. For image reconstruction, Catphan 500 was scanned, for which scans with the CT -device were performed using four different dose levels. Catphan 500 includes four modules that are used to study different characteristics of the image. In image reconstruction, SAFIRE and ADMIRE varying kernels, iteration strength and slice thicknesses. In image reconstruction, kernels are used as filters to smooth or sharpen the image. The comparison is made by examining image quality features such as image uniformity, contrast level, SNR and defining MTF-graphs. Analysis for CT -images were done using Python-based pylinac software.

The results defined image quality parameters using mean or median values for different kernels, iteration strength, dose level and slice thickness combinations. The results showed differences between ADMIRE and SAFIRE algorithms. From the results, it is found that the use of the kernel, iteration strength and dose level significantly affects the image quality characteristics. From the results, it is seen that SAFIRE and ADMIRE differ when it comes to uniformity, contrast and detecting details from the image. For example, the spatial frequency between SAFIRE and ADMIRE was approximately the same when using kernel Br38 for SAFIRE but harder kernel Br40 for ADMIRE. On the other hand, a similar SNR is obtained with a kernel Br40 using iteration strength 3 for SAFIRE but strength 2 for ADMIRE.

By studying the differences between iterative reconstruction algorithms, differences and similarities in image quality characteristics can be evaluated. When the differences between algorithms are detected, like in this work, which parameter combinations can be used for certain image quality parameters, they differ between ADMIRE and SAFIRE. Knowing the differences between these algorithms, image quality between two PET-CT scanners can be standardized and image quality optimized without increasing the dose level used in scanning.