Player is loading...

Embed

Copy embed code

Transcriptions

Note: this content has been automatically generated.
00:00:00
so uh i'm had everybody i'm sorry from matrices or building
00:00:06
basically from here um so first of all uh what is survey
00:00:11
select from your fee and white is used for and prosthetics
00:00:15
so in recent years we have seen the more and more advancements in a a technology in portable technology
00:00:22
yeah it is for seen that in the next year is
00:00:25
the solution we live translates to robotics as well
00:00:29
and to a human capabilities enhancement also for impaired people
00:00:36
however this field require the capability to the record signals from human beings
00:00:41
which uh and uh that's one euro can be useful for these
00:00:45
so what is selection menu buffy lets me your f.
00:00:47
is the measurement of electrical activity decrease muscle contractions
00:00:52
so when we thing to do a movement the movement originates into the brain any travels to target muscle
00:00:58
various are sexy we have electric chemical changes that leads to the
00:01:02
a muscle contraction that it produces an action and action potential
00:01:07
this action potential can be detected put in some lectures on this phase of the screen
00:01:13
why's election your of important for interest headaches well simply because i met
00:01:17
the muscles that control the can't actually a locating into before
00:01:21
so people that are implicated below the elbow they still had the muscles that used to control the hand before okay
00:01:31
uh lets me out of the s. b. used in hand prosthetics seems the sixties
00:01:37
uh nowadays uh eat uh has several pros and cons so nowadays
00:01:42
we have to make any uh i'd hands everybody can suffer very very advanced
00:01:46
lay mechanical point of view that can rotate the time and the released
00:01:51
that can perform up to thirty c. six a a program movements different movements of
00:01:55
the hand different press however they have also several counts so would be
00:02:00
a control c. start extremely rudimentary the training times along because
00:02:05
they require dedication of the subject to control the prophecies
00:02:08
and the uh the the alternative control strategies that have been developed are not
00:02:13
extremely usable nevertheless be a first commercial part
00:02:17
recognition systems have been recently um
00:02:21
released what is essentially take state of the art in hand prosthetics
00:02:27
um many times have been done to the control robotic and we've
00:02:32
gotta recognition movement uh classification techniques with proportional instantaneous control
00:02:38
uh with multimodal data fusion including taxidermy tourist pressure sensors or computer sensors
00:02:45
actually there are also some more complex um uh strategies that
00:02:50
in boulder for example targeting master innovation to translate the
00:02:54
uh narrows to control other mascot of the muscles um
00:02:59
that are not used for x. p. c. tasks
00:03:02
or a a neural interface is that contrast for the
00:03:05
sensation reader related to the shape and be a
00:03:09
stiffness of the object to object is objects to the uh to the person that using the processes
00:03:16
the most common approach is a a incentive researches however
00:03:21
to perform pattern recognition for movement classification would see
00:03:25
and the however in general most all all the c. stands
00:03:29
rely anyway on a similar hardware acquisition set up
00:03:32
however software setting the stage so stuffy selects from your
00:03:35
record a lexus are ways around before are here
00:03:40
the it's a lecture me over the signals heels
00:03:43
recorded it is proposes amplified filters sample
00:03:47
the segmentations perform feature extraction perform and uh finally the movement is classified for air
00:03:54
regression can be applied rather take me some of these can be applied
00:03:59
uh again into the state of the art uh we have a um
00:04:04
many publications actually so homely here that uh publications
00:04:09
uh say here till two thousand eleven or more than forty
00:04:13
and issue the deck solos control seem to be reliable
00:04:16
however uh the data ah really of
00:04:20
those publications is not publicly available
00:04:24
the acquisition methods are encouraging us or d. r. can be very diverse in different
00:04:30
publications does the results cannot be compared uh and reproduce the you see
00:04:37
so what he's uh um uh our experience he know electron
00:04:41
biography data sharing in at the state of the art
00:04:45
well i recognise that the uh uh data sharing intervals from your fee is that can give an excellent impact
00:04:51
and can be an excellent source of collaboration and that it can improve while the knowledge in the field
00:04:59
so it has been a shown in the the church aren't described in previous presentations
00:05:05
that uh a publicly available data can allow to improve different fields
00:05:10
uh includes come including computer vision medically machine and many others
00:05:14
however till two thousand eleven interface selector my your feet
00:05:17
did not have any publicly available benchmark data sets
00:05:22
so what we did into the state of the art that i described before
00:05:25
with so many publications that was what had happened to thousand leaven
00:05:30
was to create and refine are produced by the position protocol for a lecture me over three
00:05:36
profits control to up to record a publicly available data base and to realise it
00:05:42
you know the two uh allow what why tess and compare is on off classification algorithms you
00:05:48
know the to finally of mental dexterity also faces for me over the process is
00:05:52
and to reduce the uh training times so to make better prosodics that
00:05:57
these projects was uh uh funded by this this national transformations actually
00:06:02
it was led by each yep but by backup put up
00:06:05
any to this ad as part has a us and the
00:06:08
german or space centre so what was that the um
00:06:15
the acquisition protocol uh all the the of these database
00:06:19
roughly it consisted of having d. both intact and and ability subjects
00:06:25
uh uh watching some movies of movements on the screen of the laptop
00:06:31
we had several sensor located on the forum of the subjects so for amputees we add
00:06:35
uh electronic your having a minute censored that and uh
00:06:40
uh accelerometer is point of subjects we also had a separate
00:06:43
lot to record the ground truth of finger movements
00:06:47
we recorded the letterman europe the signals and we are from
00:06:51
the analysis in order to allow a robotic process control
00:06:56
those are the movements that we recorded so we
00:07:00
have a first set of uh um
00:07:03
for a set of single finger movements a set of basic
00:07:07
and increased movements so for example with quotation with
00:07:11
to be nation and a set of a grass and functional movements and also a set of course better
00:07:18
the movements where chosen by a d. a hand
00:07:22
taxonomy and hand remote expenditure sure careful
00:07:26
um the uh results so would be you know pro database actually obviously the first one
00:07:31
and deceitful because in the field but actually rather continues increase in it and it's
00:07:36
get in more more use it is publicly available it includes a six data sets currently
00:07:43
three of them are regularly is that was the other ones are waiting for a publication basically for peer review that it's quite a long process
00:07:50
it currently includes ninety seven index subjects and eleven trance radio i'm pretty subjects
00:07:55
performing approximately fifty different movements and uh uh currently we are record in
00:08:00
a a more and more data sets with different positions adopts
00:08:05
um and the database current have more than one hundred and fifty users
00:08:11
it's like two three official publications describe in
00:08:15
the database that got a uh
00:08:19
approximately a one hundred and fifty citation since two thousand and twelve
00:08:24
but obviously it's led also to several other publications related to the analysis of the results
00:08:29
it is continues an increase in the number of publications related treat because finally data are useful and
00:08:35
a number of citations because people find it useful and the easiest lead into a um
00:08:40
increasing number of international collaborations because finally people are contracting out
00:08:44
maybe to just to to recordings in the same way
00:08:49
uh it's allowed as results to test a a different data no these methods
00:08:54
so traditional classification metals deep learning methods regression with
00:08:58
the mission will then i'm ten working
00:09:01
we're paying a a way to show the possibility to uh recognise up to eleven different
00:09:07
hand movements without us a higher than ninety percent in and i'm pretty so
00:09:11
a a set of movements that could be used in real life settings let's say
00:09:16
um and the utility show the relationship between clinical parameters
00:09:22
and movements classification accuracy in a hand amputees
00:09:28
so for example we showed that body mass index uh
00:09:32
uh and that s. body mass index increase
00:09:35
the classification accuracy decreases because basically fat x. x. as insolent
00:09:40
we showed that the the longer the uh for our percentage remaining
00:09:46
big better would be the accuracy with which we can record
00:09:49
be uh we we can classify different movement which
00:09:52
is quite obvious but was not demonstrated before
00:09:56
we show that actually classification accuracy increases as
00:10:00
fun tumbling sensation increases so funny
00:10:03
sensation is the sensation of having steal your hand here once you have
00:10:08
been updated and we show that the sensation the higher the sensation be higher would be actually
00:10:14
it's which we can uh recognise the uh different movements and we show that actually
00:10:20
the longer the time last night imputation the high there
00:10:24
would be the accuracy which is a bit we're
00:10:27
and that but but is consistent somehow with the um
00:10:31
or yes we horace consistent with some results
00:10:35
a related to to to how be our cortex
00:10:39
adapt to the changes into the body
00:10:43
so what is our experience finally in a a lecture me over fee reproduce ability
00:10:51
or experiences that reproduce it really is difficult in this
00:10:55
field many parameters can strongly influence the results
00:10:59
it is fundamental to explain in detail data acquisition
00:11:03
and analysis procedures and to pay extreme
00:11:06
attention actually went reproduce in the result so when sane but you're reproducing the results
00:11:12
and it is fundamental to have benchmark data sets and proper comparison massive
00:11:18
so uh you someplace else mayor forehand prosthetics actually
00:11:23
it is fundamental to putting results that can be translated
00:11:26
to real life applications because we want amputees
00:11:30
wow one day hopefully still want to be able to control robotic and
00:11:34
as in science fiction movies uh the so depreciation setup but
00:11:39
can a influenced strongly reproduce immediately we tested in uh our database
00:11:45
uh for different acquisition setup ranging from a few hundred euros
00:11:50
to almost twenty thousand euros and uh uh um
00:11:56
actually with steamed that uh uh the classification
00:12:01
results are all ten comparable considering be
00:12:04
a senior position setup even considering the which devices and very expensive devices
00:12:10
this results uh are we going to do to to be published actually we
00:12:18
as uh since we aim to the uh um to translate a results to real life and um
00:12:24
we should we see this scene that acquisition procedure is can strongly influence reproduce ability as well
00:12:29
so for example training in uh and testing on a different
00:12:35
date acquisitions made we did seem acquisition setup but simply
00:12:39
pull it needs out and put in it again on
00:12:42
to the subjects can uh affects strongly classification accuracy
00:12:49
we have seen obviously as we all know that different
00:12:52
uh uh data analysis methods can strongly influence
00:12:56
uh the uh the the in the results of classification
00:13:00
and particular here we showed it this summer that
00:13:04
she computation or networks apply to some faceless from your face basically get
00:13:09
normal results actually so the d. d. didn't improve the results
00:13:14
that much but obviously we want to uh to to
00:13:18
the performance to continue analysing these the application of these uh two techniques to this field
00:13:26
we have seen that actually the number of
00:13:28
classes can uh strongly influence the results
00:13:32
so uh actually because the chance level varies with the number class
00:13:37
in classes and considering a specific data set feature classifier
00:13:42
the classification assurances theoretically expected to decrease as a number classes increases
00:13:49
the uh and we've seen as well that class
00:13:54
balancing then that's strongly influence reproduce ability
00:13:59
and uh so for example if but we have unbalanced classification accuracy
00:14:05
depending on the data set obviously but uh in several in some data sets
00:14:10
i mean i'm balance specification actually can be up to
00:14:12
fifteen percent higher then balanced classification at u.
00:14:17
s. c. d.'s is basically due to the fact of specific classes that uh are classified with
00:14:22
accuracy that are much higher than the other for example you know lecture my your perfect rest
00:14:28
and this is usually not pointed out in a most of the church
00:14:33
so finally in conclusion and i'll time uh the uh um a survey sites from
00:14:40
your work beforehand prosthetics state of the art so the scientific results that
00:14:45
have been shown our encouraging but the the are not easily comparable one to
00:14:51
the other they're not robust enough for real live requirements as well
00:14:56
we got a letter from your every data sharing so we really used
00:14:59
be biggest database and of first immediately routes you're the only one
00:15:05
database for surveys that american data counted include more than one to
00:15:09
one hundred subjects perform in approximately fifty different hand movements
00:15:15
he's allowed to improve public knowledge knowledge in the field and it allowed to impart
00:15:20
descended literature and to be a a and became a source of collaboration
00:15:25
we're gonna lecture me reproduce ability uh well uh we sold that many parameters uh can strongly
00:15:31
influence results so as the all within before we
00:15:35
have a uh it's difficult to reproduce
00:15:38
results uh well it is fundamental to explain the us in a publication i think to explain the
00:15:45
tape the data acquisition and all these procedures and to be extreme attention when reproducing results
00:15:52
and of others and it is fundamental as well to have benchmark
00:15:55
data sets improper comparison methods in order to allow up
00:16:00
and proper advancement in the feel that it's not like saying okay i got two percent
00:16:05
more get these five percent more but on what and the with which met

Share this talk: 


Conference Program

Welcome
Sébastien Marcel, Senior Researcher, IDIAP, Director of the Swiss Center for Biometrics Research and Testing
March 24, 2017 · 9:17 a.m.
473 views
Keynote - Reproducibility and Open Science @ EPFL
Pierre Vandergheynst, EPFL VP for Education
March 24, 2017 · 9:20 a.m.
250 views
Q&A: Keynote - Reproducibility and Open Science @ EPFL
Pierre Vandergheynst, EPFL VP for Education
March 24, 2017 · 9:54 a.m.
Reproducible Research with Bob and the BEAT Platform
André Anjos
March 24, 2017 · 10:03 a.m.
230 views
VISCERAL and Evaluation-as-a-Service
Henning Müller, prof. HES-SO Valais-Wallis (unité e-health)
March 24, 2017 · 11:35 a.m.
109 views
Q&A - VISCERAL and Evaluation-as-a-Service
Henning Müller, prof. HES-SO Valais-Wallis (unité e-health)
March 24, 2017 · 12:07 p.m.
Making experiments on remote heart-rate measurement reproducible
Guillaume Heusch
March 24, 2017 · 12:30 p.m.
149 views
138 views

Recommended talks

Accueil et présentation
Antoine Perruchoud, HES-SO Valais
Dec. 5, 2017 · 5:31 p.m.
295 views
Medical visual information retrieval: techniques & evaluation
Prof. Henning Mueller, HES-SO Sierre
Feb. 19, 2015 · 10:06 a.m.
198 views