Reference image generation

[KrisThielemans]

We currently use our "BSREM" implementation to generate the reference images. I'll show some example results here, comparing metrics with OSEM images as well.

Horizontal axis are "image updates" (often called "sub-iterations" in PET, called "iterations" in CIL).

[KrisThielemans]

Siemens mMR NEMA_IQ

Settings: 7 subsets, initial step-size 0.3, relaxation .005

objective function progress

change ylim

metrics

observations

• higher penalty strength means bigger difference with OSEM ROI values
• relaxation was too low really

[KrisThielemans]

NeuroLF Hoffman

Settings: 16 subsets, initial step-size 0.3, relaxation .01

objective function progress

change ylim

metrics

[KrisThielemans]

Vision 600 thorax

settings: 5 subsets, initial step-size 0.3, relaxation 0.01
Note: restarted after 1000 updates.

objective function progress

metrics