Thanks very much for the information Matt. Useful to know.
Unfortunately I already tried the Standard labels with our Philips data and all the components were classified as noise. We'll have to look into generating our own labels.
Any idea what the WhII dataset is?
________________________________________
From: Matt Glasser <[log in to unmask]>
Sent: 09 September 2013 11:52
To: Cole, James; [log in to unmask]
Subject: Re: Testing FIX for ICA-based resting state fMRI clean-up
I think you need to use the "Standard" training dataset. The HCP datasets
are nothing like the ones you have been acquiring (2mm isotropic, 0.72s
TR, 1200 frames, only hp=2000s, NO spatial smoothing). Thus the data are
much higher quality and we have taken a much more conservative approach to
preprocessing.
I think the temporal properties will likely make the biggest difference,
but the spatial ones are quite different too (we will have much higher
spatial frequencies available in our data).
Peace,
Matt.
On 9/9/13 11:45 AM, "James Cole" <[log in to unmask]> wrote:
>Hi Matt,
>The Siemens data are 3mm isotropic, whereas the Philips data are
>2.2x2.2x4mm. Both have a TR = 2s with 300 volumes acquired. They've had
>the same preprocessing: FSL default motion correction and high-pass
>filtering, 6mm spatial smoothing. Our group tends to limit ICA estimation
>to 25 components, though for these instances I left it unlimited to
>comply with the FIX wiki.
>Where is the WhII dataset from? And what parameters do you think are most
>important for matching between the training/labelling datasets and our
>data?
>Much obliged for the help,
>James
>
|