Player is loading...

Embed

Copy embed code

Transcriptions

Note: this content has been automatically generated.
00:00:00
right here yeah um my name is on their ranges thank you for the introduction
00:00:04
of it so um i i i was just hired a beginning of may
00:00:11
um so the stock is going to be a little bit different than the other thoughts you have seen
00:00:16
it's not going to be about is and wants it's going to be about what we can
00:00:20
be and hopefully it's a future that we're going to be able to do to get
00:00:25
okay so it's going to be about creating a buyer signal processing reputed yeah
00:00:31
i'm i'm going to explain a little bit but by the signals are
00:00:35
in which contacts we would like to uh focus our efforts on to it and
00:00:40
um briefly discuss the challenges and this would be a quick topic should
00:00:47
alright so our bodies are constantly in meeting information
00:00:52
about our health right you can be uh
00:00:55
just by looking at some by me you know that it's been a state of pain
00:00:59
or to a certain but carries a certain apologise certain genetic apology
00:01:04
or ah can also be done through uh the means of sensors right
00:01:10
so here's a few example of sensors that are commonly found in
00:01:13
the medical the main uh like trance several grams um
00:01:18
uh elect from your fee who brought a volume polls as a button was
00:01:23
using for example to measure a ground truth for these are people you work
00:01:28
a temperature heart rate ah electrical diagrams in general respiration is one so
00:01:35
okay
00:01:36
now it so happens that physicians and health care protection as in general
00:01:42
we use this values are the values measured from the systems to make treatment decisions okay
00:01:48
so it again a matter this we're gonna check with you and then they will why treatment
00:01:54
now biomedical signal or in short while signal
00:01:59
processing is the analysis of such signals to provide
00:02:03
how to how skip addictions to do that in a nice way okay
00:02:12
so basically very full the
00:02:16
if i follow the flow of dayton so the patient rate signals that are some processing and some decision
00:02:22
that can be eyes uh made by the house get production or or by the system okay
00:02:28
and we interest of course in the whole set of things
00:02:31
in this uh data flow but i'm more particularly
00:02:37
um the three less tap so the generation of the single itself the processing of the signal the refinement
00:02:43
the segmentation the understanding of that signal and the support the final decision uh there's going to help
00:02:51
the protection of to actually take the final rule we'll find decision to um up by the tree
00:02:58
so let me go fulfil examples in the area soul probably the
00:03:02
best known type of bile signals are electric audio grounds
00:03:07
everybody has already seen or been in contact with uh this kind of uh exam
00:03:12
um it's frequently used to to in house uh a house rate that
00:03:18
applications for detecting heart rate pathologies a hot related pathologies okay
00:03:25
it's until to day the most reliable way to measure your heart rate variability were holes
00:03:31
and this might be very a great it's also very cumbersome the
00:03:35
current technique today tomb uh doing c. g. is by
00:03:40
making you lie down and attach a ten different sensors on your body
00:03:46
you have to be shaved and uh the sensor going to be
00:03:50
attached and what the sensor going to measure is the um
00:03:54
electrical activity between each of the to the to censor so you can end
00:03:58
up with a bunch of one the signals that look like this
00:04:03
and it takes a very well trained um clinician to look at this and that the mine
00:04:08
if you have a problem or not okay and is of course has variations in correlations
00:04:14
we're h. uh you're in the city to your gender so it's a a difficult task
00:04:22
um another one is a letter unstoppable brown so it's a similar principle
00:04:26
this is basically trying to measure signals that emitted from your brain
00:04:31
um it's typically use to diagnose bring conditions but can be used for a number of other things
00:04:37
we've seen uh the most regions of people uh driving um video games uh using
00:04:42
e. g. um we have also seen uh people using it for identification
00:04:49
um it is a useful signal but is also very noisy okay
00:04:54
it's very cumbersome to put something that when you're had actually expect some if
00:05:01
um that was the types of bile signals such for example imagery from your
00:05:05
body could be from one next can but it could also be from
00:05:08
the phone goes off no i so the back of your mind so this is the typical accent but when you go to the ophthalmologist
00:05:14
you're basically uh they ask you to put some drops when you're when i would ask you to open
00:05:20
it and be looking site and they may take a photo and the foot would look like this
00:05:25
okay um i it's impressive but the little photo can review a lot of different pathologies in the person's
00:05:31
uh um i okay most um frequent apologies are
00:05:38
for example by beauties um michael legit degeneration
00:05:42
and um glaucoma which is a very normal disease the bomb this is
00:05:47
this is that they can cause blindness so it's a very
00:05:49
important thing to be able to the pack them as fast as possible and to have a great treatment for those okay
00:05:58
alright uh last typical example is your face itself it's also set an example about signal
00:06:05
uh the face sink into its symmetry itself can be used
00:06:09
to assist on diagnostics of this is in general okay
00:06:13
and can be also use to match the emotional state of being a
00:06:16
there was recently a paper published uh about this and that
00:06:20
so many things i don't know i counted about ten or fifteen different diseases can attack just by looking at somebody's face
00:06:28
okay so let's look if you use cases okay um the first one is
00:06:33
of course to have a machine learning and signal processing uh helping
00:06:39
helps put pictures on the diagnosis uh diseases and pathologies
00:06:45
so support for diagnostics okay so the idea is that a a particular machine learning
00:06:51
algorithm would just scan this and have the physician spot the place where
00:06:56
it should actually look at the more details okay or there's a probability of having a disease because of this particular white
00:07:03
uh the same here forty i found the same age you can run some segmentation of various
00:07:07
we can extract the veins you can't figure out that the optical disk is actually here
00:07:12
and by looking at it if it's damaged or not you can detect if the person for example is subject to ball come or not
00:07:22
now is it can imagine segmenting something like this is very costly on if you have to do it um
00:07:30
and we're gonna come back to that later
00:07:34
um so all the other use cases stella madison so the idea is that uh
00:07:42
there is um you get involved in the design off cost effective
00:07:46
devices that can actually be brought home and then a help
00:07:50
remote clinicians and to actually um on it or your status else okay so you can imagine something
00:07:57
simple searches loopholes levels but you could imagine something a little bit more sophisticated as well
00:08:03
any ideas of course that you can decrease was that's the overall cost of insurance
00:08:07
i'm not sure how many of you were here at the ice yesterday but one
00:08:11
of the talks was mentioning the we presume price of insurance in general
00:08:16
so it's an interesting thing to tackle and of course the idea is
00:08:20
to prevent fit the conditions and to work preventive lean gentle
00:08:26
um another application is sports medicine so people trying to uh twenty five performance track
00:08:31
progress portables the been more personal but can also do so professional context
00:08:38
um and if i look at the challenges at least on the brief amount of time uh
00:08:44
we started looking to the still doesn't interesting area for the up to spread on um
00:08:51
i think it would summarise them in this three categories here um the first is how far can we go
00:08:57
with a a nine to actually help physicians like knowles people or how in the agnostics from biasing
00:09:05
one gigantic problem is of course overcoming the lack of existing open they can in general
00:09:11
um so we're truly in the case here that we have a lot of small
00:09:15
data sets was more data as uh some of my colleagues have already mission
00:09:20
and the fact as well that's it you know that one of the that's data
00:09:24
is not as simple as in other areas of actually this medical doctor depend yes eight
00:09:29
this is a bad a patent for the hard in any c. g. or
00:09:34
this area here means that the person has that will come okay
00:09:38
or here is a vein and here's an at okay not my from the same age
00:09:44
so i can imagine discussed the expense of eternity the the subject is very little data so we have to deal with this
00:09:51
then there's of course the fact that you have noisy measurements ah high quality cameras
00:09:56
are available today but of course uh most of the data you have sometimes
00:10:00
dates from a very long time of also where the the quality was not
00:10:04
great there's of course a lot of these agreements some doctors would say
00:10:10
this person is likely to develop a particular disease some others would
00:10:14
say against okay and of course population biases we see a lot of uh challenges nowadays
00:10:20
in the area of health going i'm open uh there was one recently me fine
00:10:26
about the technical palmer from my phone those images which is very interesting but uh sometimes um
00:10:34
this challenges they don't take in consideration for example your new c. d. or your age
00:10:38
and the analysis of those as understand now differs depending on this categories before
00:10:47
well then i would like to also work a bit on improving sensing okay for
00:10:51
diagnostics we're thinking about where oppose but we also have in house expertise
00:10:57
emotive spectrum multi device signal position that we can probably leverage as well
00:11:03
and work on early diagnostics so how can you prevent of really ah
00:11:08
estimate that somebody's about for example to have a hard stroke or to develop ah
00:11:14
i'm michael or degeneration for example in the back of the
00:11:18
okay so
00:11:21
the question is why tibia up okay so the idea here is that
00:11:25
we try as much as possible and it's interesting uh platform because
00:11:29
there's a lot off expertise existing in complimentary area okay
00:11:35
i've annotated here few but this is by
00:11:38
no means exclusive as a sign the c. d. and e. g.
00:11:42
there's a lot of sleep is processing the noisy required ah
00:11:46
we're going to need of course new a. i. too so do very good in shops segmentation of these
00:11:51
images measure different parameters but we also gonna be you willing to work with us small data sets
00:11:58
so this is a quite um difficult problem as already
00:12:01
um what four by david ah better sensing technologies
00:12:07
to be able to um actually see where is difficult to see
00:12:13
we're going to be have to be handling as well uh problems with sort of dayton privacy as
00:12:18
you know house your personal house is private data about metrics but it's probably even more uh
00:12:26
sensitive than that because it's talking about your pathologies would be of diseases nobody really wants to get that in your
00:12:33
connections the prosthetic that work as well um the control of robotics uh
00:12:39
fact that interact for example with the electron my ram actually control the what the hand i think is also point that
00:12:47
would uh work altogether and houses of course uh social science
00:12:52
before there is another set of connections between um
00:12:57
this study of health in general social computing and even understanding in natural
00:13:02
language process it as as already mentioned by some of my colleagues
00:13:07
now it yeah that is a great place to stop is this it has had a policy and pointing is our
00:13:13
policy of open softer and opened it does so in forcing open science which i think is very good
00:13:18
and that we see more more but is still lacking uh probably in house
00:13:24
connections export connections of labs working with sensors without exactly having bill how on how to do the
00:13:30
analysis of the data so some of it's it's some of the exist are all this year
00:13:36
in house expertise in complimentary areas as i just mentioned excellent support them system development project
00:13:42
management infections or you think you have already talked to some of these people
00:13:48
and uh uh some inspections for
00:13:52
couple of uh upcoming uh initiatives you
00:13:57
i see and you know it's okay
00:14:01
so the ambition is that yep joint of their domain was the work of this group
00:14:06
vi it's applied the i know how and it becomes instead of artificial
00:14:11
intelligence for society our division intelligence for how to sew side
00:14:18
right so and my top here i put here my list of interest to make contact information
00:14:23
if you haven't checked out or posters outside by asking to do so to do in the coffee break
00:14:29
um there are two that i'm particularly the you're a route off for bob is uh to get four principal research that we
00:14:37
intend to continue to use in this new group and or be put form which is the but from open signs

Share this talk: 


Conference Program

Introduction by Hervé Bourlard
BOURLARD, Hervé, Idiap Director, EPFL Full Professor
Aug. 29, 2018 · 9:03 a.m.
916 views
Presentation of the «Speech & Audio Processing» research group
MAGIMAI DOSS, Mathew, Idiap Senior Researcher
Aug. 29, 2018 · 9:22 a.m.
16949 views
Presentation of the «Robot Learning & Interaction» research group
CALINON, Sylvain, Idiap Senior Researcher
Aug. 29, 2018 · 9:43 a.m.
9721 views
Presentation of the «Machine Learning» research group
FLEURET, François, Idiap Senior Researcher, EPFL Maître d'enseignement et de recherche
Aug. 29, 2018 · 10:04 a.m.
14111 views
Presentation of the «Uncertainty Quantification and Optimal Design» research group
GINSBOURGER, David, Idiap Senior Researcher, Bern Titular Professor
Aug. 29, 2018 · 11:05 a.m.
3210 views
Presentation of the «Perception and Activity Understanding» research group
ODOBEZ, Jean-Marc, Idiap Senior Researcher, EPFL Maître d'enseignement et de recherche
Aug. 29, 2018 · 11:24 a.m.
5620 views
Presentation of the «Computational Bioimaging» research group
LIEBLING, Michael, Idiap Senior Researcher, UC Santa Barbara Adjunct Professor
Aug. 29, 2018 · 11:45 a.m.
4130 views
Presentation of the «Natural Language Understanding» research group
HENDERSON, James, Idiap Senior Researcher
Aug. 29, 2018 · 2:03 p.m.
8976 views
Presentation of the «Biometrics Security and Privacy» research group
MARCEL, Sébastien, Idiap Senior Researcher
Aug. 29, 2018 · 2:19 p.m.
6512 views
Presentation of the «Biosignal Processing» research group
RABELLO DOS ANJOS, André, Idiap Researcher
Aug. 29, 2018 · 2:43 p.m.
4027 views
Presentation of the «Social Computing» research group
GATICA-PEREZ, Daniel, Idiap Senior Researcher, EPFL Adjunct Professor
Aug. 29, 2018 · 2:59 p.m.
7158 views

Recommended talks

L’analyse de données et ses bénéfices pour le patient
Immo Oliver Paul, fondateur de Carenet (Brésil)
June 6, 2014 · 1:03 p.m.
Management of health data in an industrial company
V. Srikanth Nallanthighal
Sept. 5, 2019 · 1:57 p.m.
142 views