Congress:
EuroSafe Imaging 2021
Keywords:
Artificial Intelligence, Radiation physics, Radioprotection / Radiation dose, Conventional radiography, CT, Image manipulation / Reconstruction, Observer performance, Physics, Radiation safety, Quality assurance
Authors:
C. Hoeschen, G. Frija, K. Katsari, D. Koff, J. Damilakis, R. W. Loose, J. Santos, M. van Straten, I. Kralik, F. Kainberger, M. Kupinski, D. Tsetis, C. Granata, S. Schindera, D. Tack, P. Hogg
DOI:
10.26044/esi2021/ESI-15558
Methods or background
There are various approaches to try to cope with these demands:- one can perform receiver operating studies (like in 1) or other studies looking for subjective image quality descriptors based on imaging tasks. The big challenge in this case would be to have a sufficient subtle ground truth to detect deficiencies in methods. Besides of this, such studies are very time consuming and need a lot of effort to achieve relevant results.
Looking for physics based, objective approaches to determine image quality- one can look for task-based approaches in general addressing detection tasks for various backgrounds, again related to dedicated imaging tasks (like described in general in 2 or for specific questions in 3 and 4).- one could try to measure Fourier-based image quality parameters directly in patient images (like in 5 and 6).- some easy system characterization might be feasible with Fourier-based image quality measures in phantom images. However, that would only be feasible for linear imaging approaches. This would for sure exclude the characterisation of the performance of iterative reconstruction methods and even more of AI based reconstruction in CT for example.
It is necessary to correlate the outcome of the above mentioned measures to diagnostic performance, since that is the relevant parameter in the optimisation of the patient's benefit / risk ratio. This is even more important and difficult for the objective image quality assessment approaches than for the subjective approach.