Which denoise process? pros and cons.
Dear Dipy developers and users, I am planning to preprocess DWI data to study free-water in both white matter and grey matter. I have notice that there are at least 3 differnt denoising options (NLMEANS, local PCA and Marcenko-Pasteur PCA) and I would like to kindly ask (whoever wants to answer) , which are the pros and cons of each one, or which one is most commonly used by default. Thanks in advance Dani Bergé Hospital del Mar, Barcelona, Spain
Hi Dani, Sorry for the slowness on our end. In my opinion, Marchenko-Pastur PCA should be used as the default. 2 things to ensure, 1. Keep patch_radius parameter = 2 by default. If you get an ill-conditioned error, increase the patch radius. 2. Marchenko-Pastur PCA is considered to be a non-aggressive denoiser, meaning that it leaves in some noise at the cost of not removing any signal. You can follow this tutorial - https://dipy.org/documentation/1.1.1./examples_built/denoise_mppca/#example-... where we show the effect of denoising on DKI parameter maps. You should be able to see a similar reduction in degeneracies in FW-DTI (if any) due to noise suppression. If the noise in your data is too high, I would go for the empirical Local PCA/ NLMeans. Where you will need to play around with the sigma parameter a bit. Let us know how this goes or if you need any more help on our end! Regards, Shreyas ________________________________________ From: Dani Bergé <dbergeba@gmail.com> Sent: Wednesday, May 6, 2020 7:29 AM To: dipy@python.org Subject: [External] [Dipy] Which denoise process? pros and cons. This message was sent from a non-IU address. Please exercise caution when clicking links or opening attachments from external sources. ------- Dear Dipy developers and users, I am planning to preprocess DWI data to study free-water in both white matter and grey matter. I have notice that there are at least 3 differnt denoising options (NLMEANS, local PCA and Marcenko-Pasteur PCA) and I would like to kindly ask (whoever wants to answer) , which are the pros and cons of each one, or which one is most commonly used by default. Thanks in advance Dani Bergé Hospital del Mar, Barcelona, Spain _______________________________________________ Dipy mailing list -- dipy@python.org To unsubscribe send an email to dipy-leave@python.org https://mail.python.org/mailman3/lists/dipy.python.org/ Member address: shfadn@iu.edu
Hi Dani, Sorry for the slowness on our end. In my opinion, Marchenko-Pastur PCA should be used as the default. 2 things to ensure, 1. Keep patch_radius parameter = 2 by default. If you get an ill-conditioned error, increase the patch radius. 2. Marchenko-Pastur PCA is considered to be a non-aggressive denoiser, meaning that it leaves in some noise at the cost of not removing any signal. You can follow this tutorial - https://dipy.org/documentation/1.1.1./examples_built/denoise_mppca/#example-... where we show the effect of denoising on DKI parameter maps. You should be able to see a similar reduction in degeneracies in FW-DTI (if any) due to noise suppression. If the noise in your data is too high, I would go for the empirical Local PCA/ NLMeans. Where you will need to play around with the sigma parameter a bit. Let us know how this goes or if you need any more help on our end! Regards, Shreyas ________________________________________ From: Dani Bergé <dbergeba@gmail.com> Sent: Wednesday, May 6, 2020 7:29 AM To: dipy@python.org Subject: [External] [Dipy] Which denoise process? pros and cons. This message was sent from a non-IU address. Please exercise caution when clicking links or opening attachments from external sources. ------- Dear Dipy developers and users, I am planning to preprocess DWI data to study free-water in both white matter and grey matter. I have notice that there are at least 3 differnt denoising options (NLMEANS, local PCA and Marcenko-Pasteur PCA) and I would like to kindly ask (whoever wants to answer) , which are the pros and cons of each one, or which one is most commonly used by default. Thanks in advance Dani Bergé Hospital del Mar, Barcelona, Spain _______________________________________________ Dipy mailing list -- dipy@python.org To unsubscribe send an email to dipy-leave@python.org https://mail.python.org/mailman3/lists/dipy.python.org/ Member address: shfadn@iu.edu
participants (2)
-
Dani Bergé -
Fadnavis, Shreyas Sanjeev