Hi Paulo!
Thanks a lot for the link (www.mint-labs.com). I didn't know about your
company but it looks like you're running a really wicked project! I'd
love to test the beta for the 3d visualization.
I registered to mint-labs but I couldn't see how I can download your
data (I'm also interesting in machine learning myself) - is there maybe
some link I didn't check?
I also noticed that you apply your mlearning techniques to other
neurologically interesting exams, such as EEG, MEG and PET: that's
fantastic.
I'm aware that in the ward where I'm doing my internship as an undergrad
surgeons tend to use a C-Arm (something like this
http://usa.healthcare.siemens.com/clinical-specialities/surgery/surgical-disciplines/neurosurgery)
for both spinal and brain surgery: for example in the case of some
aggressive brain tumor, such as a blastoma, a common combo would be
C-Arm + Surgical Navigation System (something like this
http://www.medtronic.com/for-healthcare-professionals/products-therapies/neurological/surgical-navigation-and-imaging/neurosurgery-imaging-and-surgical-navigation/
, which essentially is a supercool GPS where the Earth is the brain and
the hand of the surgeon are like satellites) + 5-aminolevulinic acid
(for fluorescence guide surgery) + microscope (of course).
Unfortunately, pretty often, the 5-ala is not able to impregnate well
enough the totality of the cancerous cells, which of course means that
the surgeon can't distinguish them from the healthy tissue and,
consequentially, not being able to remove them all, their presence could
result in a tumoral relapse.
Considering that both the microscope (live) and the C-Arm (whenever the
surgeons uses it) return an imagine with potentially a decent
resolution, do you think it would be mad to try to attempt to see how
some convolutional neural network would perform against 5-ala in
detecting healthy and cancerous tissue?
My idea would be this:
- CNNs/RNNs --> to analyze the bare "screenshot" (or frames) coming from
the microscope and the C-Arm;
- NuPIC "GPS" System --> to be adapted for the coordinates given by the
Surgical Navigation System and trying to predict where potentially the
tumors "goes" (certain tumors have really nasty shapes and there are
really tiny little cellular lines that squeeze into distant part of the
brain - same thing goes for certain spinal tumors).
Would this seem possible to do?
I'm sorry if I had to give these kind of basic explanations but I did it
in order to keep the topic interesting and relevant for other people who
may read this email that maybe are not familiar with the topic :)
Thanks,
Raf
On 15/01/2016 20:42, Paulo Rodrigues wrote:
Hi.
Yes, MRI (and especially brain MRI) provides 'static' images - either
showing the structure as gray matter or white matter (like from
diffusion MRI) or other modalities.
But if you want a time series on it I would say fMRI is more
interesting, as it measures 'brain activity' as a proxy from the blood
oxygenations, i.e. if part of the brain is working more, it is
consuming more energy, so it needs more blood. fMRI is discussable, or
better the conclusion that are drawn from it, especially in the task
based studies.
Now what is particularly of interest and cool is resting state fMRI
<https://en.wikipedia.org/wiki/Resting_state_fMRI> - the subject is
not doing any task, simply resting in the scanner, and the brain is
'fluctuating' in it's "idle mode". And then you can do cool stuff like
look at the brain as a network, and as a dynamic network. This is
being actively researched as a predictor for pathologies, cognition,
etc etc - and it also has some nice connections with philosophy :) It
has difficulties in the possibilities of analysis methods and
pre-processing it, but depending on your brother's research needs,
something can be done more or less complex.
We have bunch of data if you need. Just let me know.
Have a great weekend!
Paulo
ᐧ
On Fri, Jan 15, 2016 at 8:26 PM, Matthew Taylor <[email protected]
<mailto:[email protected]>> wrote:
Mark, Marion, or Kentaro might have something to say abou this.
Also, here are some examples of similar projects:
-
https://www.youtube.com/watch?v=gzfTZhd6X9c&index=12&list=PL3yXMgtrZmDqZc2m7qI3Kkbmxechp2-Zs
-
https://www.youtube.com/watch?v=Ij4StdJBxEo&index=17&list=PL3yXMgtrZmDqZc2m7qI3Kkbmxechp2-Zs
---------
Matt Taylor
OS Community Flag-Bearer
Numenta
On Thu, Jan 14, 2016 at 1:09 PM, gideon isaac
<[email protected] <mailto:[email protected]>> wrote:
I was looking at some of the audio applications of Nupic
provided by a helpful person on this list, and I think the
same methods could be used for an MRI application.
Static MRI pictures are constructed from Fourier transforms of
a signal produced by Hydrogen atoms spinning under the
influence of MRI magnets. They are then converted into a
picture. So temporal is converted to spatial or vice versa.
I’m somewhat muddled on this, but my brother creates sequences
for MRI machines and then gathers the data coming out, and I
want to interest him in Nupic.
MRI static pictures are usually used to look for anomalies
like tumors, or soft tissue problems that do not show up on
x-rays. It might be hard to compare a pathology picture with
a healthy picture because people vary in height and shape
etc. That could be a problem – there is no standard picture
of a lung for instance – you would have to feed pictures from
people of all sizes into the HTM.
There would be no need for a step of converting the signal to
a picture, instead the signal would be fed directly into Nupic.
There is also Functional-MRI – which can watch movements – you
can watch a MRI movie of a heart beating and look for
anomalies there.
So is this something I can present to my brother’s research
team at his university, and if so, are there any special
methods that should be used?
Thanks.
--
*Paulo Rodrigues, PhD
CEO & co-founder
*/Arc de Sant Silvestre 4, entresuelo segunda
08003 Barcelona, Spain
Tel. +34 933 282 007
Mob. +34 633 817 514
/[email protected] <mailto:[email protected]>
www.mint-labs.com <http://www.mint-labs.com>
--
Raf
www.madraf.com/algotrading
reply to: [email protected]
skype: algotrading_madraf
