Optimization of dose in relation to image quality in synchrotron based tomographic lung-imaging

Abstract: Introduction: Lung imaging with synchrotron light sources is currently a widely studied field. The image resolution and quality are remarkable and the level of detail possible to study the smallest structures, the alveoli, of the lungs cannot be obtained by any other non-invasive imaging modality with the current technology. The most common and useful type of lung imaging with brilliant light sources is tomographic imaging which is being performed exclusively on rodents at the current state. However, in the case of tomographic imaging with synchrotron light sources the absorbed doses to the samples are high due to the high photon flux. Although exposure times don’t extend a few milliseconds the number of irradiated angles is normally in the order of 900. Thus, the main objective was to investigate whether there is an optimal number of projection angles that should be used in order to maximize image quality while simultaneously minimizing the dose to the sample i.e. maximize the image quality-to-dose ratio. A subobjective of the study was to investigate the improvement of image quality when generating phase contrast images compared to standard absorption contrast images. Materials and Methods: A tomographic acquisition of static mouse lungs were performed at the TOMCAT beamline of SLS. In the frameworks of Python, the data was reconstructed with all projections (900) using two different reconstruction algorithms, Gridrec and SIRT. For the investigation of the subobjective the data was reconstructed, by both algorithms, with and without the application of a phase retrieval algorithm. The main investigation was done by generating 8 images with each reconstruction algorithm. The images were generated with a decreasing number of projection angles. The first included all the projection angles, the second kept only every other projection, the third kept every third etc. Thus, the number of used projection angles for the 8 images were 900, 450, 300, 225, 180, 150, 128 and 112 respectively. ESD was used as dose concept and CNR as a measure of image quality. The theoretical ESD was calculated for the images with different number of projection angles and CNR. Finally, the CNR-ESD ratio was determined. Results: As Kitchen et. al. (2017) suggests it could be seen that the image quality was greatly improved with the application of phase retrieval to generate phase contrast images in comparison to standard ACimages. This was the case when using both the Gridrec and SIRT algorithms for reconstruction as the