<div dir="ltr"><div><div>Dear Lotte,<br></div>Thanks for the data. I have had a look (you can thank Ana for the reminder). The snapshots show indeed a bad result. My advice is to work on a small example first, as the one
enclosed for this test (you can create one from real data to). Since I have never worked with SIRT before, I first checked SIRT's definition and let's agree on this paper <a href="http://dx.doi.org/10.1109/TMI.2008.923696" target="_blank">dx.doi.org/10.1109/TMI.2008.<wbr>923696</a> (which I recommend if you want to use SIRT). Then I noticed that it does not update after the first iteration and, indeed, the last subset was not used which gets unnoticed with SART but not with SIRT... The fix is in a new branch named sirt and will be merged later today if all tests pass. The results look more as expected, i.e., no convergence after 100 iterations but something closer to the solution. Check the mentionned reference for improving convergence (they suggest 1.99 as a relaxation parameter) but my feeling is that the conjugate gradient algorithm is more efficient (and it's been more tested in RTK).</div><div>Best regards,</div><div>Simon<br></div><div><br></div></div>