Dear Krzysztof,

Have you tried playing with some of the datasets at openneuro.org? They're not specifically designed to have a certain amount of noise, but could still be valuable as a real-life testing set since they do feature real-life MRI data, obviously including scanner noise.

Of course you can also use them as a base to manually add noise to (e.g., using ricernd() ) and then test your method.

All the best,

Johannes

Dear Johannes,

Thank you for your answer. It is an interesting source of data.

One more question. Is it methodologically correct adding noise (using ricernd) to a real-life MR image?

Krzysztof

If you want images with the greatest impact for Rician noise I would recommend you focus on the DTI (Diffusion).

Dear Krzystof

Are you interested with a test data with high b-values weighted diffusion MRI ?

matrix size of 196×196×16 (field of view=2.7cm×2cm×1.2cm)

native resolution 0.137755102 x 0.102040816 x 0.75 mm

31 b-values :

6 independent directions

As the b-value increase the signal decrease , then the SNR decrease but with 31x6 equivalent estimations of the noise baseline level

Images of animals. One hemisphere with high SNR, the contralateral has less signal, in the same image (pathological model).

You should have the noisy to ultra noisy image you want to deal with.