[Music]
>> So my name is Ivan Tarapov.
I work for Microsoft Research.
Our team is called InnerEye.
This is the project InnerEye,
and we work on the medical image
analysis.
And what I'm showing here is the
tool that we built which is a
tool that is used by radiation
oncologists during cancer
treatment planning.
So what happens in cancer
treatment planning is that a
trained physician, a radiation
oncologist, would have to look
at a 3-D medical image of a
patient.
So we're looking at a typical
computer tomography scan that
they would take, and they would
go through every slice of this
image and they would accurately
delineate all the organs that
are interesting to them.
The organs that have the tumor,
but also the organs that need to
be spared due to the treatment,
because eventually those
delineations will go into this
radiation therapy machine which
effectively needs to focus a
high-energy beam of ionizing
radiation of the tumor and try
to have it shrink or destroyed,
but at the same time it needs to
spare everything that's around
the tumor.
So what they do right now, they
would basically go through every
slice and they would do
something like that.
They would say that I'm fine to
contour bladder here, right, and
on this slice I'm going to draw
a contour like that.
And, by the way, you may notice
that I'm using the Surface
Studio and a pen, which they
don't have, so they would use
the mouse and keyboard, and then
they would switch to the next
slice and draw another contour
here and then they will switch
to another slice, which gets
pretty laborious and takes a lot
of time.
But there's tool in the market
to improve this proces, but we
believe we have built something
that allows them to be even more
efficient.
So we have this fancy button
here that says add a structure
set.
And when I click on it, I have
this selection of organs that
I'm telling my algorithm to look
for.
So I'm pressing start.
And what happens right now is
machine learning algorithm is
running in the background, going
through every single voxel on
this image, trying to classify
it, tell whether this voxel
belongs to a bladder or to a
prostate, or it's the left femur
or right femur.
So it's pretty efficient.
This machine here has an I7
model CPU, and it takes us only
under a minute to come up with
the segmentation.
So here we have the full
segmentation of a prostate CT
scan.
So it's not 100% accurate, but
So it's not 100% accurate, but we have all the organs in there
we have all the organs in there
and all the contours are very
and all the contours are very precise and all that's there's
precise and all that's there's
left to do is, for example,
here, where we have the prostate
bladder interface, probably
didn't get it right, I would
click the add/remove and I would
clean up those pieces of
prostate segmentation that went
into here, and rather than
contouring 110 slices, you
would have to go through
probably 10 and fix them up and
the rest -- and that gets you
there.
Also in this tool we have this
nice volume render which helps
you visualize what's in there in
3-D.
Yes, so this is it.
This is Project InnerEye.
There's a lot of things that we
plan to do, a lot of things you
can do with machine learning
based medical image analysis.
And so we're working with other
regions and other problems in
radiation oncology and radiology
alike.
So for now this tool is
available to select clinical
institutions.
But if you're one of them, and
if you're curious to help us and
participate in this program and
get your hands on this tool,
please contact us, Project
InnerEye, and thanks for your
attention.





Không có nhận xét nào:
Đăng nhận xét