THIS IS HISTORY! Check fabien.benetou.fr for news.

Seedea, scalable creativity

Xye, consultancy for serious creators

Information

RSS

(Updates)

back to the Fundamental concepts tree project

Discussion on imaging techniques

Principle

  1. you write down a model of how your mind is working (let's say you think that you think that way, that you use causation, correlation and compare, etc...)
  2. to validate or invalidate your model you use fMRI (or another technique) to check that during a specific action (let's say tasting 2 different kind of chocolate) your brain actually does what you think it does
    1. if it does, perfect, it confirms your hypothesis
    2. if it does not then you have to work on it again (going back to step 1)

Result

Techniques

NameMRIfMRIDTIEEGPETCTMEGOptogeneticsDSIfNIR
TR 3s       microseconds
Spatial resolution   poor     5mm
Invasiveness low lowlow    no
Cost         depends on channels but 808nm LED
Usage          
SoftwarePyNIfTIPyMVPA
3D_Slicer
Camino
TrackVistriathlon    TrackVis 

Usable products

Prices are out of date, check the dedicated Wikipedia page.

NameEmotivNeuroSky    
TechniqueEEGEEG    
Price$500Private
(NDA required)		    
API/SDKEmotiv SDKMindKit-EM    
CommunityFB group     

See in general Wikipedia:Comparison of consumer brain-computer interface devices

Discussion

Explanation on fMRI and their limitation

[08:54] M: what you see in fMRI is activation of different zones in the brain.... zones that are already known... i mean, they haven't discovered anything.. they've just seen that what they already knew was correct [08:54] M: let me show you an example [08:54] Fabien: yes activation zone, but also patterns of activations right ? [08:55] M: epilpeptic patiens don't have a normal brain... their brain has evolved in a different way than that of a healthy patient [08:55] Fabien: yes [08:56] M: lets assume that our patients has a node in the speach area of the brain.... and he was born this way [08:57] M: ok.. the brain is plastic, that means that it will try to rearrange itself to work at 100% [08:57] M: so if the epileptic node is in the "standard" speach zone, the "new" speach zone will be somewhere else [08:57] M: ok? [08:58] M: so fMRI comes in because i can see where is this zone in my patient [08:59] M: the ultimate need is to fully locate the node, and be able to surgically remove it.... without affecting brain performance [08:59] M: i don't think you could see paterns of activation... [09:00] M: for example, wiht certain images, you activate your visual cortex and maybe your memory, or even the word associated to that image... [09:00] M: but in fMRI you might see every zone activated.... you will not see that one area is activated first or second [09:01] M: MRI works adquiring images in a time window, TR... [09:01] M: standar TR for fMRI is 3s [09:02] M: in 3 seconds, it makes 30 (i.e) scans of your brain, achieving a volume.... [09:02] M: but you can see that response times in the brain is a lot faster ( A LOT ) than 3 seconds [09:03] Fabien: (reading) [09:03] Fabien: ok [09:04] Fabien: so maybe not fMRI actually, maybe another technic is better [09:04] M: fMRI is quite new... i think it can evolve a lot [09:04] Fabien: to me fMRI = live study of the brain, ability to see disctinct zone activation with short time span, find ordering patterns, etc [09:05] M: by patterns you mean: i see a drawing, i realise it's an apple, it reminds me of the taste? [09:06] M: and the areas involved? [09:07] Fabien: that would mean recognition of a known object (function) = see, find, compare, remind (pattern) [09:07] Fabien: with each see, find, compare, remind activating different zone [09:07] Fabien: so with a temporal study you can see a repetitive pattern [09:08] Fabien: each time you make someone look at a well known object [09:08] Fabien: the same areas will "light up" [09:09] M: yeah, but that was already seen... [09:09] M: it's a known fact... [09:10] Fabien: ok [09:10] Fabien: so then [09:10] M: that's why you can compare epileptic patients to regular people [09:10] Fabien: yes yes

Example

Let's imagine that you want to be smarter so that you can make nicer present to a girl you like and not mess it up each and every time. You think :

Then instead of just thinking about it you write it down on a sheet of paper :

Then you smile and you think :

You then decide to use an fMRI or another technic to actually try to match your model to cold hard data. You design an experimental protocol and find that your model was kind of correct but not perfect. You refine it, change it and try it again. Still no perfect, so you do it again (perfect in the sense of realistic, accurate, not ideal). And you keep on repeating the process till you are satisfied.

Nature of the written model

[09:16] M: what you write in paper is which areas of your brain should be working? [09:16] Fabien: no no [09:16] Fabien: it's on the mind functions, not the brain area [09:17] Fabien: like a language of your mind, not the positions themselves [09:17] Fabien: that would be too hypothetical now I guess (but should check the state of the art in neurology, maybe now it's possible) [09:17] Fabien: you write something like recognition of a known object (function) = see, find, compare, remind (pattern) [09:17] Fabien: and you write functions like that one [09:18] Fabien: for see, find, compare, remind too [09:18] Fabien: it is just a model ok, it is probably not correct [09:18] M: so you would do it to build models of thinking? [09:18] Fabien: yes exactly [09:19] M: and apply it to a.i. [09:19] Fabien: well [09:19] Fabien: eventually of course [09:19] Fabien: but even without [09:20] Fabien: let's imagine that now you have a model of thinking [09:20] Fabien: and that based on your experiment it is accurate [09:20] Fabien: that's good [09:20] Fabien: but you want more, of course [09:20] M: why not? [09:20] Fabien: yes, if you can and worked hard for it, why not being able to do better

Going further than realistic modelling

[09:21] Fabien: so you have your model of thinking [09:21] Fabien: and because of your ability to think about it now [09:21] Fabien: you can try to improve it ! [09:22] Fabien: but just the paper version... [09:22] Fabien: which is nice but ... not really practical [09:22] Fabien: BUT [09:22] M: yes... and i guess that it could be the other way around.... meaning that if you find out that the paper version is better, you could train yourself to follow that line of thinking [09:22] Fabien: yes, you got it [09:23] Fabien: and I think fMRI too can help for it [09:23] Fabien: because now you have an aim to reach [09:23] Fabien: you know that you should light up this zone then that one etc [09:23] Fabien: you have live visual feedback and for learning I think this is fundamental. [09:24] M: yes, but i think fMRI should improve timing issues to be able to work like that [09:24] M: 3 seconds is a lot of time [09:24] Fabien: yes [09:24] Fabien: I thought it was much much faster

Alternatives to fMRI

[09:24] Fabien: I heard about another technic [09:24] Fabien: that is less precise regarding locations [09:24] Fabien: but more precise regarding time span [09:25] M: which one? [09:25] Fabien: ... I can't remember the name, Ive seen it in a documentary months ago... [09:26] M: you could make it faster, but you loose info... [09:26] Fabien: but it is not really that important, what do you think about this idea, makes sense to you ? [09:26] M: we use a TR of 2.901 seconds... [09:26] Fabien: that's way too slow [09:26] M: yep.. for what you want, yes... but for us its fine [09:27] M: most fMRI studies consists of blocks of activations of 30s and deactivation of 10-15seconds [09:27] M: so then you compare activation vs. non-activation [09:27] M: so... tr of 2.901 s works just fine [09:28] Fabien: I see [09:28] Fabien: it wasn't a criticism of the tool [09:28] M: i know i know [09:29] Fabien: so, from what you know about neurology, the tools, learning, etc does it make sense to you? Is it crazy but still "possible" somehow? [09:29] M: errr... yeeahhh... maybe it's possible... [09:30] Fabien: because the result is pretty interesting. [09:31] M: check Diffusion tensor imaging [09:32] Fabien: impressive, Ill have to check this out and make a comparison sheet of the different technics [09:32] Fabien: thanks a lot [09:33] Fabien: (the images are beautiful :) [09:33] M: yes... i don't know a lot of this... but i think it shows "highways" of info in the brain [09:33] M: http://www.nmr.mgh.harvard.edu/nouchinelab/pdfs/DaSilva%20et%20al2003.pdf [09:34] M: i have a friend that is working on (diffusion tensor) though...

End of a discussion on #openvibe on freenode

20:41 <@K3rl0u4rn> et bien fais nous signe lorsque tu passes à Rennes et nous verrons si nous pouvons dégager un créneau ;)
20:41 <@K3rl0u4rn> je me sauve
20:41 <@K3rl0u4rn> a+ tard
20:41 < Utopiah> ca marche, bonne soiree
20:41 <@K3rl0u4rn> merci de même
20:41 -!- K3rl0u4rn [n=yrenard@lns bzn-38-82-253-86-168.adsl.proxad.net] has quit ["Leaving."]

then again 01/07/2010 (cf logs) 

14:51  Asu-> je peux rien garantir mais il devrait y avoir du monde oui =)
14:51 < Asu-> ça dépend des congés de chacun

Resources to explore

Page last modified on July 12, 2010, at 10:08 AM