Dear Diane,
The proper statistical way to distinguish meaningful activation from
noise is to perform a statistical comparison either between conditions
or between activation and baseline period. Once you export something
to images SPM doesn't care whether the images were derived from single
trials or from multiple trials. Obviously for single trials SNR will
be much lower so your effects will have to be very strong or you will
need a lot of trials to find something.
Technically there is no difference between performing a source
reconstruction of an ERP and performing source reconstruction of a
single trial. You just need to either generate a series of datasets
with one trial each or take an epoched dataset and mark all the trials
as bad except one at a time.
However, I think a more promising approach for most purposes will be
to assume that the same set of brain sources is activated for each
trial and the only difference is in degree of activation. You can do
that by assigning a different condition label to each trial (something
like 'stim1', 'stim2', 'stim3' etc.) Then you invert all the trials
together but you'll be able to export images for each trial
separately.
Best,
Vladimir
On Wed, Jul 29, 2009 at 9:18 AM, Diane Whitmer<[log in to unmask]> wrote:
> Dear Karl,
> Another theoretical question for you: Is it possible to perform source
> reconstructions of the induced EEG response on a single-trial basis? What
> would be the proper (statistical) way to distinguish meaningful activations
> from noise?
> Thank you,
> Diane
>
> ---------- Forwarded message ----------
> From: Giulio Ruffini <[log in to unmask]>
> Date: Fri, Jul 17, 2009 at 3:23 PM
> Subject: Fwd: Source reconstruction of induced activity in SPM
> To: Diane Whitmer <[log in to unmask]>
>
>
>
>
> Begin forwarded message:
>
> From: Karl Friston <[log in to unmask]>
> Date: July 17, 2009 2:15:16 PM GMT+02:00
> To: Vladimir Litvak <[log in to unmask]>, "Rik Henson"
> <[log in to unmask]>
> Cc: Iván Cester <[log in to unmask]>, Bernhard Spitzer
> <[log in to unmask]>, Nicholas Furl <[log in to unmask]>,
> Giulio Ruffini <[log in to unmask]>
> Subject: Re: Fwd: Source reconstruction of induced activity in SPM
> Dear Vladimir and Diane,
>
> There is nothing written about this (although perhaps we/you should perform
> a systematic
> analysis of a well-characterized induced response). The rationale is not
> based
> on the forward model or its inversion; it is based on feature-section. In
> other words,
> you need to select the data features you want to reconstruct and make
> inference about
> (this is an area called active learning in the machine leaning community)
>
> In this case, the mechanism is quite transparent, To endure precise
> conditional
> estimators of source activity the data are projected onto the first few
> eigenvectors
> of the temporal covariance. This subspace is the feature space and can be
> controlled
> by band-pass filtering. If you want to look at evoked responses look at
> alpha frequencies;
> if you want to reconstruct gamma, then band pass to perverse gamma.
>
> There are no rules for feature selection; it may take a few months before we
> know
> how to do this properly. Remember that feature selection is a little bit
> like experimental design - it
> allows you to focus on or harness those aspects of data that constrain the
> unknowns most
> effectively.
>
> I hope this helps,
>
> With very best wishes,
>
> Karl
>
>
>
> At 10:28 17/07/2009, you wrote:
>
> Dear Karl,
>
> Could you give Diane a more theoretical answer than I've given? I'm
> afraid my theoretical depth has been exhausted.
>
> Thanks,
> Vladimir
>
> ---------- Forwarded message ----------
> From: Diane Whitmer <[log in to unmask]>
> Date: Fri, 17 Jul 2009 11:11:44 +0200
> Subject: Re: Source reconstruction of induced activity in SPM
> To: Vladimir Litvak <[log in to unmask]>
> Cc: Iván Cester <[log in to unmask]>, Bernhard Spitzer
> <[log in to unmask]>, Nicholas Furl
> <[log in to unmask]>, Giulio Ruffini <[log in to unmask]>
>
> Hi Vladimir,
> I'd actually like to understand on a theoretical/mathematical level why the
> algorithm will only work for bandpass filtered signals. Is there a paper you
> could point me to? More specifically, why does it not work to find a
> solution for all frequencies, and then bandpass afterwards?
> Thank you,
> -Diane
>
>
> On Wed, Jul 15, 2009 at 4:02 PM, Vladimir Litvak
> <[log in to unmask]>wrote:
>
>> The algorithm will model features in the data and if your feature of
>> interest is in particular frequency band you should specifically look
>> at that band. Otherwise if for instance you are interested in gamma
>> but there are much stronger low-frequency components in your data, the
>> algorithm will model those components and not gamma.
>>
>> Best,
>>
>> Vladimir
>>
>> On Wed, Jul 15, 2009 at 2:57 PM, Diane Whitmer<[log in to unmask]>
>> wrote:
>> > Thank you for the software, Vladimir.
>> >
>> > Could you please explain why it is that the algorithm requires band-pass
>> > filtering?
>> >
>> >
>> >
>> > 2009/7/10 Vladimir Litvak <[log in to unmask]>
>> >>
>> >> I forgot to reiterate that you should pre-filter your data around the
>> >> frequency of interest.
>> >>
>> >> Vladimir
>> >
>> >
>> >
>> > --
>> > Diane Whitmer,
>> > Researcher
>> >
>> > C. de l' Observatori Fabra s/n
>> > 08035 - Barcelona - Spain
>> > tel. +34 93 254 03 68
>> > fax. +34 93 212 64 45
>> > skype. dwhitmer
>> > email. [log in to unmask]
>> > ______________________________________
>> > STARLAB, Living Science http://www.starlab.es
>> >
>> > This e-mail and the attached files contain confidential information and
>> are
>> > of exclusive use for the addressee. The disclosure, copy and
>> > distribution
>> of
>> > the present message and its attachments are prohibited. If you are not
>> the
>> > intended recipient and you receive this communication, we inform you
>> > that
>> it
>> > is forbidden by law to read, copy or use it. Please notify us at
>> > [log in to unmask] and proceed immediately to delete it.
>> >
>>
>
>
>
> --
> Diane Whitmer,
> Researcher
>
> C. de l' Observatori Fabra s/n
> 08035 - Barcelona - Spain
> tel. +34 93 254 03 68
> fax. +34 93 212 64 45
> skype. dwhitmer
> email. [log in to unmask]
> ______________________________________
> STARLAB, Living Science http://www.starlab.es
>
> This e-mail and the attached files contain confidential information and are
> of exclusive use for the addressee. The disclosure, copy and distribution of
> the present message and its attachments are prohibited. If you are not the
> intended recipient and you receive this communication, we inform you that it
> is forbidden by law to read, copy or use it. Please notify us at
> [log in to unmask] and proceed immediately to delete it.
>
> Kind regards,
> Giulio
>
>
> Giulio Ruffini
> CEO, Starlab
> C. de l’Observatori Fabra s/n - 08035 – Barcelona - Spain
> tel. +34 93 254 03 62 /66,
> mob. +34 679 49 57 77
> fax. +34 93 212 64 45
> e-mail. [log in to unmask]
> ________________________________________
> STARLAB, Living Science http://starlab.es
>
> This e-mail and the attached files contain confidential information and are
> of exclusive use for the addressee. The disclosure, copy and distribution of
> the present message and its attachments are prohibited. If you are not the
> intended recipient and you receive this communication, we inform you that it
> is forbidden by law to read copy or use it. Please notify us at
> [log in to unmask] and proceed immediately to delete it.
>
> Este correo electrónico y los archivos adjuntos al mismo [en su caso]
> contienen información confidencial y de uso exclusivo para su destinatario o
> destinatarios. Queda prohibida la divulgación, copia o distribución de la
> información contenida en el mismo y, en su caso, de la información contenida
> en los archivos que pueda llevar adjuntos. Si usted recibe este correo por
> error, no siendo el destinatario del mismo, le indicamos que por ley no le
> está permitido la lectura, copia y uso del mismo. Se ruega lo notifique lo
> antes posible a la dirección [log in to unmask] y proceda a borrar el mensaje
> inmediatamente después.
>
>
>
>
>
>
>
>
>
>
> --
> Diane Whitmer,
> Researcher
>
> C. de l' Observatori Fabra s/n
> 08035 - Barcelona - Spain
> tel. +34 93 254 03 68
> fax. +34 93 212 64 45
> skype. dwhitmer
> email. [log in to unmask]
> ______________________________________
> STARLAB, Living Science http://www.starlab.es
>
> This e-mail and the attached files contain confidential information and are
> of exclusive use for the addressee. The disclosure, copy and distribution of
> the present message and its attachments are prohibited. If you are not the
> intended recipient and you receive this communication, we inform you that it
> is forbidden by law to read, copy or use it. Please notify us at
> [log in to unmask] and proceed immediately to delete it.
>
|