prabhatkc / ct-recon

Python (n some matlab) Implementation of Forward & Inverse models for biomedical imaging
9 stars 0 forks source link

Is the noise addition repeated? #2

Closed Love-Sunshine closed 1 year ago

Love-Sunshine commented 1 year ago

Thank you very much for the code provided, but I have identified several minor issues. In file insert_noise_2_obj_model.m, have both pct_ldsino and yuct_ldsino added noise repeatedly? Why is k_nd determined to be 0.85? The ndsino_n in im(3, ndsino_n, 'sino'), cbar; also does not appear in the code. How can these problems be solved? Thank you very much for your answer!

prabhatkc commented 1 year ago

Firstly, I have fixed the plotting issues in the file, insert_noise_2_obj_model.m and updated GitHub with the corresponding fixes. Secondly, yes, both the functions, pct_ldsino and yuct_ldsino, add noise. pct_ldsino yields normal-dose acquisition (which still contains some level of noise), and then yuct_ldsino further adds noise to acquisition from normal-dose to yield low-dose measurement. Refer to the references cited in the header part in the same Matlab file. Next, k_nd=0.85 was heuristically determined to match the NPS curve obtained from the uniform scan provided in the low-dose grand challenge dataset. Look at the part https://github.com/prabhatkc/ct-recon/tree/main/error_analysis/cho_lcd#nps-validation-against-ldgc to get more idea on this aspect. Finally, I have uploaded a demo file, https://github.com/prabhatkc/ct-recon/blob/main/digiNoise/demo_m1_validation.m, with a noiseless reconstruction example.
I hope this info helps.

Love-Sunshine commented 1 year ago

Thanks a lot, I can understand!