Embed code
Note: this content has been automatically generated.
00:00:00
gonna make everyone so i'm feeling quite privilege here to show a very
00:00:05
technical or a work because when compared to do the presentation
00:00:08
of this morning a feed him much behind in terms of vision
00:00:12
and platforms there are four four rip releasable which uh research
00:00:16
but the message on to convey here on at least for this domain
00:00:18
of american image analysis is that you can have advancement does
00:00:22
but the only moment where you know if it works on that is really when you run
00:00:27
it on lots of of cases and and then you can do is rip releasable research
00:00:32
so what i'd like to to show of today is a medical image
00:00:36
analysis method for prices i met scene and more in particular
00:00:39
the three d. risk of iron spectrum of that that uh i would
00:00:43
for this cried and then show about france validation on the
00:00:46
small a set of a of a round uh the big concern human city
00:00:51
and then how we can we could build an open access ready mix what but found
00:00:55
that the doctors can use to further very dated with the with the on patients
00:01:01
so this is the outline i would first introduced the the motivation which
00:01:04
is to do non invasive personalise estimation of concept written success
00:01:08
and the whole the methods how to get there was text operators integration functions
00:01:13
then this validation on this might get a little customer that assets and then the platforms punching with these conclusions
00:01:20
some education's behind this work is that we know that if you have a c. t. image of of it you more
00:01:27
that the the structure of this uh do you want to see it is you
00:01:31
reflects the natures for instance you can see does out different uh i didn't consume us in c. t.
00:01:36
you have regions which are very dense these are regions are rather active cancer cells
00:01:40
are all the regions otherwise under under changes et cetera so this is
00:01:44
very uh had originally was it's not enough to in that we don't you
00:01:48
fry those regions but it's also important to understand the eco system
00:01:51
of the different regions because they are like seven michael habitats that are
00:01:56
uh working in completing which of the the beach with each other
00:02:00
and if we are able to understand heterogeneity then valuable to crack and divide the type of some of cancer
00:02:06
and also there for two to predict if you can if the treatment can success on
00:02:12
so this is uh what we like to do when he's walking this is more generally regrouped in under the terms of of radio mix
00:02:19
and what we want to do here in particular is to extract a
00:02:22
treat extra features which are related to the morphological properties of tissue
00:02:26
and uh to do that to predict response to a rigid therapy basically so they
00:02:31
process is like that's you you have that you more here the native country
00:02:35
we extract the set of features and then we hope for instance that
00:02:38
uh the the these patients that we'd have a treatment success for even came with the puppy
00:02:44
will populate a similar region of the features base which would help us basic it we don't divide them and say okay
00:02:49
justin based on the image than intensively uh being a good to say that this the treatment would work on that
00:02:55
so this activity surrogates uh the current uh person meeting based on
00:02:59
biopsies where you go with a needle you take a piece
00:03:02
of tissue which is quite critical if you're in the lines are in the brain and then you run very uh yea
00:03:08
along the agenda make analyses and you end up with with on with with the with not come
00:03:13
and also because based on the major have an overview of that you more okay
00:03:17
we don't have the molecular precision that we have an overview of the heterogeneity
00:03:22
so of course there are lots of walking does the main uh included going does
00:03:27
but they are back to main limitations that would like to highlight here is that the first one
00:03:32
is that there is no separate and then id analysis of the node components so basically they always
00:03:36
a average the the the the measures of the entire to more than the for you
00:03:41
you mixing like or have it that's another uh some limitation is that usually the size
00:03:47
of this vector here is usually larger than the number of points you have in
00:03:50
the space and we all know that is really leading us meaning to just went
00:03:54
over fit so this is to aspects we'd like to address in this work
00:03:59
so that's the decorator and and see how we can uh use like local measures
00:04:04
of of text you're here and for that we define the general uh
00:04:08
framework for texture which considers that n. e. g. dimensional uh takes genesis
00:04:13
approach can be characterised by a set of local operators g. n.
00:04:18
and also part also look out in the sense that their response so if
00:04:21
you apply to separate out what particular position x. x. zero hindering image
00:04:26
the response only depends on this a small box on the small special support here
00:04:31
do you have an operator like that you have an image like that thing and to apply the separate from this image to extract properties locally
00:04:37
so for instance uh if you do that with the now operate toss
00:04:41
you have like the compulsion you we'll go a conversion operators
00:04:45
and so basically come all this image uh to you you that's got product of this guy
00:04:48
who is is this image at every position you end up with a response map
00:04:52
which would highlight specific project properties for instance uh for for for that you know you not textual that you
00:04:59
so in this work we use a a one particular type of operator which as our it's way let's
00:05:06
and basically the these guys they they work a bit like a direction that image theory
00:05:11
that you said so you can see that because this is the collection of of
00:05:15
text operator was that gives you now defining for you so you need to play
00:05:19
by jeremy guy here in the numeric toss again you get there that is
00:05:22
that are like somewhat normalised but the vector of this uh and we get here so you have kind of or bus image directories
00:05:29
and so they have an intuitive but interpretation because for instance we take all the one you you get the slope to
00:05:34
the gradients of the transition between the pixels although to is the is the the the the curve it's up there
00:05:42
so if you look at them when you can hold it with this is a
00:05:45
something that is band pass it looks like that's we have basically three d.
00:05:49
with that second order very directly them next wise et cetera and um
00:05:55
actually there is nice property with them is that uh there's terrible which means that we act
00:05:59
actually uh can express and rotations of them within our condition of the bases sediment
00:06:05
and which help us to activate locally i line and those with that's for each position to make
00:06:10
sure that we are normalised orientation we are locally rotation valiant of for for the images because
00:06:15
we expect to be able to describe is look at properties within vines to edition
00:06:22
okay so now we define local operate us and we'd we'd like to have
00:06:26
the aggregate them over region basically instance that you more so we
00:06:30
we apply the operators to all position in the image and then we we get the response now
00:06:35
but in yeah we want to have one ms you for that you more not a set of of of response everywhere
00:06:40
so we need to aggregates those measures uh to estimate feature statistics and we need to do that or region of interest
00:06:47
that could be for instance the core of that you more all the margin of two more on anything else so
00:06:53
to do the segregation there sort options natural one is to do integration and the the
00:06:57
most the simplest example is with average would you do the average of each
00:07:01
operator responds uh over that you also get a set of features here
00:07:06
uh but there is a very strong limitation was average as as we know it
00:07:11
might give some signal so for instance i like to do straight something here
00:07:15
so imagine we have this text you're here which was clearly two distinct a visual events that's it
00:07:21
and then we have to operate also job just like secretary symmetric here they just look
00:07:25
at things of difference gets here so we can see the response maps over here
00:07:30
and now we will aggregates this is a response or different regions here so we can aggregate
00:07:36
over the the right the uh all the green ones over or over the whole image
00:07:40
and you could look at the feature space that this stands so if we look at the the average
00:07:44
response of of just operate or here on the x. axis and separate on the y. axes
00:07:49
we can see that all the green regions are very well uh regrouped same thing for the red ones
00:07:55
and when you average over the want entire image with the blue one end up or something in me the rejected
00:08:00
doesn't correspond to anything does your average this is this is not happening this image that's a big problem
00:08:06
so coming back to our program of of an aside to mars this is
00:08:09
the problem because if you average picture properties over these heterogeneous two more
00:08:14
then you end up with something that doesn't correspond to him sing biological essentially get your signal and and you cannot do anything with
00:08:22
so let's see another way here to drug regulation which is still based on summarising the
00:08:27
the response of the boat also over region but men using the the covariance matrix
00:08:33
so now it's a bit better because we keep for each position the uh the
00:08:38
cool actually visions of the of the features if you want so we have a set of operators and we
00:08:43
know they the covariance together that's that's a given point so this is what we will use here
00:08:48
and then you can for instance the vector rice them here to to obtain actually if you recognise your components metrics
00:08:54
of just to go per triangular because they're symmetric you end up with with a with a vector like that
00:09:00
now we need to take into account the facts does does mattresses
00:09:04
we cannot just measure that couldn't distances between these vectors
00:09:07
because they leave on the symmetric positive this and if you need a menu for no
00:09:12
so it should take a a naked in distance between them this is wrong it's a
00:09:15
bit imagine you have to drive a from now that one out for instance that
00:09:19
it that's useful to know the the couldn't distance between these two points because you
00:09:23
cannot fly between them you need to take the role to go there
00:09:26
so this is the same saying you needs to try don't put this money
00:09:30
for the company that couldn't distances between these two guys not helpful
00:09:34
so they are when using distance that that exist if you want to compute
00:09:39
the distance between these two koreans but excellent going to do this here
00:09:42
and what if we want to do more than computing distance but to compute scallop
00:09:46
products for instance to do a c. m. classify our our feature or anything
00:09:51
well uh it's also possible that but you have to uh look at me
00:09:55
uh approximate the properties of human before so basically you take one production point which is
00:10:00
uh uh between the two a manual so i see this fear is is the menu full of the covariance matrices
00:10:06
and this is a production point and then you can uh project all your points
00:10:10
on this plane tangent here with what we would call logarithm exponential metrics operations
00:10:15
that's the point on this point here so i won't go into details but you you need to know that we can go from this menu
00:10:21
forced to displaying and on this plane we can do uh and it's
00:10:26
got a product so we can compute as yams for instance
00:10:29
and you can have your decision functions that taking to a account
00:10:33
like one man and a covariance matrix and the train hype
00:10:38
then of course to stop products depend on the production points so and it's computed like that
00:10:44
alright so these other but that's that that we have that we have and let's see how it works on a on a i dunno customer
00:10:51
so what we wanted to do here is we have a ninety to run like
00:10:56
uh the customer from uh our colleagues that lets them for hospitals and clinics
00:11:00
for each of these uh cases we have uh the regions donated
00:11:04
to have the grass to mow volume should entire remote
00:11:07
the grandest capacity which is the room want excluding the what one and sorry
00:11:12
regions which are the most familiar news which are the red ones
00:11:16
so we will use the different regions recognition to see if it makes a difference and also we have the disease free survival
00:11:23
time so which means that we know that this is t. it was zero and then they started the register putrid
00:11:29
and then that detracts which patients have the relapse during this period
00:11:33
not will do in this war could just simply a binary thing so we will consider all the patients
00:11:38
that's how the relapse before twelve months he before one year or not okay so
00:11:45
this is our that binary amount that we want you to uh to predict
00:11:49
now we we estimated generalisation performance using the telephone 'cause edition
00:11:53
for each training fall we need to estimate this projection point for the menu full
00:11:57
of basically will use the minutes of the koreans of the training set
00:12:00
uh so that it's cool to everyone and you will of course train after as yams on this on this tension plane
00:12:07
for each test for that we will then compute the classification groups so these are the results
00:12:12
so we fixed some parameters of our framework we repeated the the experiment several times
00:12:17
we have a here quite and buttons that the sets because you have twenty three recurrence within the for twelve mourns next
00:12:23
to sixty nine or missions and these are the the results so that makes bandwidth what we have in the stables
00:12:30
i i is when we use the full framework so using the s. yams a trained on the menu fault
00:12:36
b. is when we use the covariance matrices but we compute okay then distances uh
00:12:42
between them since actually you can fly between uh the and one of the
00:12:46
and each line here is when you agree getting over is the ontario to
00:12:50
most of the g. t. v. sorted in red or did you
00:12:54
that's mainly focused on the f. one score because this isn't balanced so so uh this
00:12:59
is a better uh ms your for in buttons that asset if one score
00:13:02
and we can see that already just using the the fact that this covariance leave on the on the
00:13:08
the symmetry defeats the many fall we can already improve a little bit to the the performance
00:13:14
uh and especially when we're gonna get over the entire to go for some seventy six to eighty
00:13:21
that's such an also interesting to see what kind of results we gets when we
00:13:24
do would simply do average over the of that you or that you more
00:13:28
and then we we see that this is a big problem because here when you have a range of the
00:13:32
anti r. two more the g. t. v. uh we get the really big drop in terms of performance
00:13:37
and this is a big plus the case it should just to get them on the certain parts because those guys every how much uh news so it's
00:13:42
still okay to agree get an homage is regions of course but the prime
00:13:45
happens is when you do an average over over a hydrogen you see
00:13:51
so i just want to our bodily had happy so it seems to work as we want
00:13:56
but it's still only ninety two patients with a number features just is about the same
00:14:01
so we we don't know if this so this is all on your feet on
00:14:04
so this is why we are no building a a here an open access when you mix with by phone
00:14:10
corrected it talks like that so you can uh go to this website you can test it
00:14:15
and that then you for the doctor last they can just a big zip file with
00:14:19
all their passions and then they can have like a c. team age even pets
00:14:23
and and the uh the region of interest you can submits and it would process them in the batch months we
00:14:28
can go uh we with hundreds of patients you you choose a few parameters packet textures and things like that
00:14:34
then you can get results like uh in this easy file that you can plug
00:14:39
into your father it's a statistical software lex is p. s. s. or are
00:14:43
and also like basic ways to to miss your uh to use your statistics on the on the features
00:14:49
so we we started to use just got from now to further validate what we've
00:14:53
done and really to your hands to the doctors that they can ah
00:14:57
a pro that's of patients at this and see if if it works for for
00:15:00
the problem we already started to have a very interesting results uh on that
00:15:04
it's also nice for me because i finished my job i i we program that and then that's it's the
00:15:08
doctors that need to find it it it and and and and get some people sort of it is
00:15:14
okay so it's time to go to conclusion that button perspectives and so we were able to predict written failures
00:15:20
uh within the first and pull phones for isn't gonna go see of of eighty percent so this is
00:15:25
but useful for doctors they can choose if you want to use the the treatment or not um you've
00:15:30
seen that is covariance a free market elegance to agree getting it without really killing the information basically
00:15:36
and we have beats this up and that's us a web platform of l. that is really
00:15:40
available now publicly also for last evaluation so we are instilled have dogs because okay
00:15:46
some people named differently there five since it was we were still into this phase of
00:15:50
making it work for for most people but uh i think we're getting there
00:15:54
and and the switch or work then we really want to to buy those red you mix uh features
00:15:59
for for being using pick a routine so so this needs a lot of patience like
00:16:03
it's you have hundred features we would like to have thousand patients seem like
00:16:07
so thank you very much for for for attention

Share this talk: 


Conference program

Welcome
Sébastien Marcel, Senior Researcher, IDIAP, Director of the Swiss Center for Biometrics Research and Testing
24 March 2017 · 9:17 a.m.
Keynote - Reproducibility and Open Science @ EPFL
Pierre Vandergheynst, EPFL VP for Education
24 March 2017 · 9:20 a.m.
Q&A: Keynote - Reproducibility and Open Science @ EPFL
Pierre Vandergheynst, EPFL VP for Education
24 March 2017 · 9:54 a.m.
VISCERAL and Evaluation-as-a-Service
Henning Müller, prof. HES-SO Valais-Wallis (unité e-health)
24 March 2017 · 11:35 a.m.
Q&A - VISCERAL and Evaluation-as-a-Service
Henning Müller, prof. HES-SO Valais-Wallis (unité e-health)
24 March 2017 · 12:07 p.m.

Recommended talks

Projet GOGE : Comment établir, gérer et évaluer un business model ?
Daniel Amrein, HES-SO Valais-Wallis
20 May 2014 · 6:15 p.m.