 |
 |
Hi Tina, I would say the more directions, the less sensitive to noise results would be. There are lots of paper in the literature that describe optimal acquisition protocols for DTI, in general I would go for at least 20 directions, unless certain time limitations exist. Cheers, Stam On 7 Mar 2011, at 14:44, Tina Li wrote: > Hi Stam, > So it depends on how many diffusion direction we get, the more directions, the better results. Can I say it this way? > > Thanks, > Tina > > At 2011-03-07 22:35:54，"Stamatios Sotiropoulos" <[log in to unmask]> wrote: > Hi Tina, > > dtifit uses least squares to fit the model to "noisy" data and returns point estimates. As I mentioned before, we usually have more datapoints than unknowns and that is a way to decrease the uncertainty of the estimates (think about fitting a line to 2 noisy points and fitting it to 10 noisy points, you will be more confident with the second fit). > > Hope this is clearer, > Stam > > > On 7 Mar 2011, at 14:25, Tina Li wrote: > >> Hi Stam, >> I am sorry that I have not explained clearly. As we know that, the noise is distributed in both brain and non-brain region, the non-brain area is masked, so we can remove that part of noise, but there is still noise in the brain area. My question is that how the dtifit do with the noise inside the brain? So as you explaination,dtifit do not deal with the noise inside the brain. Am I right? >> >> Thanks, >> Tina >> >> At 2011-03-07 21:40:13，"Stamatios Sotiropoulos" <[log in to unmask]> wrote: >> Hi Tina, >> >> I am not sure what is the question here, do you mean whether we do any prefiltering? The answer would be no, we come around the fact that diffusion images are noisy, by using more diffusion samples than the number of unknowns in the DTI model. So in general you would acquire 20-30 diffusion directions, while you have 7 unknows to estimate. >> >> Cheers, >> Stam >> >> >> On 7 Mar 2011, at 09:11, Tina Li wrote: >> >>> Hi, Stam >>> >>> Yes, I believe you are right about DTIstudio. >>> In FSL,the non-brian area is avoided by mask, what about the noise in brain area? >>> >>> Thank you, >>> Tina >>> >>> At 2011-03-07 16:56:21，"Stamatios Sotiropoulos" <[log in to unmask]> wrote: >>> >>> Hi Tina, >>> >>> I do not know how DTIstudio works, but I believe this is an intensity threshold and not SNR threshold, i.e. voxels with low intensity are considered backround, outside the brain. In FSL, BET is used to extract a brain mask. Estimation in non-brain areas is avoided by applying this brain mask when running dtifit. >>> >>> Cheers, >>> Stam >>> >>> ----- Original Message ----- >>> From: Tina Li >>> To: [log in to unmask] >>> Sent: Monday, March 07, 2011 8:48 AM >>> Subject: [FSL] DTIFIT and noise >>> >>> Hi, FSL expert >>> When I use DTIstudio to get the FA and tensor, there is a window where I can put in a value to move the background noise, for example, if I put in a value 30, then will remove all the value below 30. >>> I wonder how the FSL control the SNR when I run ditfit to get FA and tensor? >>> >>> Thanks, >>> Tina >>> >>> >>> >>> >> >> >> > > >