Phase data seem to be ignored in CLI

paquiteau / patch-denoising

Patch based denoising method, targeting fMRI thermal noise removal.

https://paquiteau.github.io/patch-denoising/

MIT License

21 stars 4 forks source link

Phase data seem to be ignored in CLI #8

Open tsalo opened 1 month ago

tsalo commented 1 month ago

The phase data are loaded, but then immediately the variables are overwritten with magnitude data:

https://github.com/paquiteau/patch-denoising/blob/b34360bbfe62f6c2db320cd0c5523faf3f493ea8/src/patch_denoise/bindings/cli.py#L129-L131

Also, it looks like only the magnitude noise map is supported, since there's just a --noise-map parameter but no --phase-noise-map parameter. Are the phase noise volumes not used by NORDIC?

paquiteau commented 1 month ago

well, this is a bug. Several things to do:

add an "else" between 130 and 131
add similar support for noise map. However in the context of 0-volt acquisitions, I think we don't need the phase information, and we could use the Rayleight Distribution ^1 to estimate back the variance of real and imaginary part (assuming they are iid)

tsalo commented 1 month ago

Is there any reason not to use the phase noise volumes if they're available? Or are you saying they're not strictly necessary if you don't have them, but you would use them when you do have them?

paquiteau commented 1 month ago

If you have phase data, use it , if not we can use the Rayleigh model (or just assume it's gaussian, for high enough snr (> 10 iirc) its okay)

tsalo commented 1 month ago

Ah, that makes sense. Thanks!