Number of iterations and its relation to the noise level and smoothness

General Discussions
1

Hello Everyone,

I have reconstructed an image with 24 iterations. However, when I look at the profile of the 3rd iteration and 24th iteration, I see a discrepancy between the noise level and the smoothness of the two images. See the attached screenshot for details. Is it right that by increasing the number of iterations, we get spiky profiles? Probably there is an empirical threshold for the number of iteration, right?

I used the following command in the terminal:

castor-recon -df test.cdh -opti MLEM -it 24:7 -proj joseph -conv gaussian,.5,.5,3::psf -dim 240,240,240 -vox 0.4,0.4,0.4 -dout output

Iteration 3:

it3
it32560×1318 40.6 KB

Iteration 24:

it24
it242560×1314 43.8 KB

Thanks,
Seyyed

2

Hi Seyyed,
I don’t see anything wrong from your profiles, the variance will increase with the number of iterations. As for the “right” number of iterations, it is complicated to give an answer as it depends on various elements such as your data, system, reconstruction algorithm,… or even the purpose of your reconstruction (eg, qualitative or quantitative study). Usually the number of iterations/sub-iterations are chosen in order to get a trade-off between accuracy, precision and computing time.
Best,
Thibaut