Related, how important are FEAT's intensity modification steps prior to the melodic process?
________________________________________
From: Christian F. Beckmann [[log in to unmask]]
Sent: Thursday, December 10, 2009 5:28 PM
To: [log in to unmask]
Subject: Re: Melodic Steps
Hi
Multiple differences:
- You're not using any variance-normalisation - even with the vn switched off melodic does calculate some relevant quantities (but doesn;t perform the VN scaling then)
- 2a works differently for multiple reasons, amongst others, of computational complexity
- 2b does not remove the mean again - otherwise you might actually remove important spatial signal signatures
hth
Christian
On 9 Dec 2009, at 23:48, Srinivas wrote:
> Hi,
>
> I am implementing temporal concatentation approach of Melodic in MATLAB
> and trying to follow the steps as mentioned in the melodic documentation.
>
> 1. Data - Read data for each subject and arrange in time points by voxels.
> Time series mean is removed per voxel.
>
> 2. a. First PPCA - Stack data of all subjects spatially and compute eigen
> vectors. Project data of each subject onto the common eigen space
> (Whitening).
>
> b. Second PPCA - Remove mean of timeseries per voxel for each subject's
> reduced data. Stack all subjects data temporally and compute eigen vectors.
> Project stacked data on to the common eigen space.
>
> 3. PICA - Reduced data from the second PPCA step is used to compute
> component spatial maps and time courses. Spatial maps thus obtained are
> indepenent.
>
> After doing step 2b., group PCA map in my code is not matching the spatial
> map obtained using Melodic. Please let me know if I am missing some steps.
>
>
> Thanks,
> Srinivas
|