Speech Transcript - Trends in Bio Imaging and Signal Processing

0:00:13	it nice but the
0:00:14	yeah speech can kind of thing but choose by see image and signal processing
0:00:19	and we have here and
0:00:20	for as the have this can who twenty
0:00:24	in
0:00:25	oh
0:00:25	and that are the hot topics
0:00:29	and two oh so as to the community at large of signal processing
0:00:35	so um
0:00:37	a for the yeah i
0:00:40	and that the
0:00:42	uh about a
0:00:45	gene
0:00:46	is to be a a
0:00:47	and
0:00:48	first that which is a the perspective of the scene or the best G
0:00:52	to understand the can the kind for five
0:00:56	uh_huh
0:00:59	hmmm
0:01:03	a i like the
0:01:07	and the prediction some genes and proteins and a is to be able to it
0:01:11	is is the conditions
0:01:13	so
0:01:14	and the prior to choose
0:01:16	but G
0:01:19	two
0:01:21	the band
0:01:22	and the size of a
0:01:23	down to the size of a
0:01:26	the the content was a scale and by different that from for example a pad this is medical imaging
0:01:34	the the the some of P be to go just to be
0:01:41	she's which i produce a information which just produced
0:01:44	we have a number of topics which i uh i have uh come to the two
0:01:50	and discipline
0:01:51	which i a
0:01:53	to to this process and and then said
0:01:56	good friend mathematical gym which is how can put
0:01:59	some i think so that's processing
0:02:02	i
0:02:06	uh
0:02:07	uh
0:02:09	uh
0:02:10	i information a set uh but and then
0:02:15	to to of us steps
0:02:18	a
0:02:18	first
0:02:19	processing knows those problems
0:02:23	can can and then
0:02:26	to a a a a time of just first we choose to put into a just uh a just a
0:02:32	images which type of crime
0:02:35	so that the the fact to extract information from the background the relevant information
0:02:41	to the classification of information
0:02:43	and the to be able to put in in um
0:02:50	true
0:02:51	in this manner under images are kind of
0:02:55	in this manner as
0:02:59	and and uh i and in and then
0:03:03	no side then i
0:03:05	hi
0:03:05	this
0:03:07	i just to uh for size
0:03:11	and the fact that for methods
0:03:14	and
0:03:15	for the processing session
0:03:18	uh for the
0:03:21	to look
0:03:23	a can to the to the information
0:03:27	to choose so uh uh uh a type is for example
0:03:33	compressed sensing is becoming
0:03:35	a of research
0:03:37	but had to go
0:03:39	because the the
0:03:41	filter
0:03:44	i
0:03:46	to to spend to much time
0:03:50	this
0:03:51	and a a which is how does from
0:03:56	it it's also about uh
0:03:58	and
0:03:59	and
0:03:59	oh
0:04:01	um
0:04:01	type of research
0:04:04	okay
0:04:05	uh
0:04:07	so so i could use a set have two
0:04:11	so that
0:04:12	was which i X uh
0:04:15	yeah
0:04:16	but
0:04:18	to
0:04:20	that
0:04:22	to test uh
0:04:26	i
0:04:28	a uh a a uh and each of them
0:04:32	so so as to have this
0:04:34	was
0:04:39	um
0:04:40	a process right
0:04:43	which the channel
0:04:44	have to the go to a a a uh of the um a
0:04:51	as the beaches
0:04:54	so
0:04:56	yeah uh i use and try and
0:05:01	okay
0:05:03	can give some idea is also at could me find find
0:05:08	G Q
0:05:14	and
0:05:23	uh_huh
0:05:25	i'm can some stuff
0:05:27	so that's think that's that uh
0:05:31	so are so T R screen D that's as far as E and a go to note
0:05:38	some
0:05:40	access from shown
0:05:42	and information and uh
0:05:47	so
0:05:48	and
0:05:51	my first slide okay
0:05:56	and T
0:05:58	because i asked to at some found notion and that function as a cat and so i don't you can
0:06:05	still
0:06:07	so
0:06:09	i have
0:06:10	um
0:06:11	i have a sounds
0:06:13	have stop to learn and then sit products and i've that's that's you know
0:06:21	and and so
0:06:25	i
0:06:25	that
0:06:27	so that so that
0:06:29	asked
0:06:31	uh_huh
0:06:32	asked
0:06:34	and
0:06:36	so
0:06:37	and then suppressed
0:06:40	so um
0:06:42	and that X M M M and that's consistent
0:06:51	that's as a
0:06:52	so that's that's so
0:06:55	five
0:07:00	and so
0:07:02	and at the i-th sensor and this is yeah
0:07:09	that sounds
0:07:10	i
0:07:12	a uh
0:07:16	thanks to that construction
0:07:19	five so
0:07:21	so i don't function and sabrina
0:07:25	and uh i know that have to reconstruct
0:07:30	i have
0:07:32	i
0:07:34	yeah
0:07:37	the presence of a
0:07:39	so here
0:07:42	and that's that's faq
0:07:49	so that's accent
0:07:52	i have to say that the construction masters channel
0:07:58	i
0:07:59	uh
0:08:01	i don't
0:08:02	i have a pension plan
0:08:06	oh doesn't that
0:08:10	so um
0:08:12	the acquisition system
0:08:16	and and twenty seven are expressed as a function
0:08:23	and at times i was
0:08:27	of uh uh
0:08:29	optimization is convex and how
0:08:33	a a a a a a and uh some kind of
0:08:38	and uh
0:08:40	or section
0:08:43	i
0:08:44	you
0:08:46	a and have and it's fast but was that because we've got a a a a a a three D
0:08:51	the mountains to kind
0:08:56	press sound system so it's down
0:09:02	so a five a script
0:09:06	and oh oh can script
0:09:10	yeah that's a
0:09:16	um
0:09:19	uh i i stands as a facts phones
0:09:24	i have some
0:09:26	so
0:09:27	and and that's a that's it's function
0:09:31	how
0:09:33	that that
0:09:35	that's
0:09:37	facts phone so so that yeah
0:09:41	the and that and a and the convolution "'cause" of times perhaps function as as nine
0:09:47	that's a concepts
0:09:51	and
0:09:54	but
0:09:55	a she used on the point spread function
0:09:58	so that an absent system a
0:10:03	so account the mouth
0:10:06	i have just stash um
0:10:11	a a i have a complex to that so that the structure i
0:10:19	and uh that's a constant
0:10:24	uh i
0:10:27	yeah
0:10:29	i was just a concept for example of god
0:10:34	construct estimate costs
0:10:36	for five example some kind of cats watson's mention that
0:10:40	uh
0:10:44	as a last step to
0:10:48	for some sequences come out
0:10:52	match well
0:11:01	um
0:11:03	i
0:11:04	i
0:11:05	hello
0:11:06	yeah
0:11:07	and so on
0:11:09	have come to that
0:11:12	to the concept of a complex when the structure
0:11:16	that's that's a problem
0:11:18	oh have
0:11:21	uh statistical stuff such a complex task
0:11:26	and five is not uh
0:11:28	to define approach to insist substance and not rocks
0:11:33	and uh have questions that is the kind of news and most of them to a convent but sectors
0:11:41	in some sense
0:11:43	yeah
0:11:48	we propose is that we
0:11:50	uh_huh go through an addition that the have ten minutes for discussion open discussion on all the the the topic
0:11:58	so yeah also go on was biological imaging
0:12:02	and perhaps web addresses a little for being very focused some biological imaging because this was also andrew
0:12:08	was
0:12:09	the supposed to to about medical imaging that in
0:12:13	uh uh in through a right now
0:12:15	i
0:12:16	so yeah
0:12:18	uh
0:12:18	hmmm and it is so was so as so he of the challenge signal processing challenges for
0:12:24	microscope
0:12:25	okay
0:12:26	and uh uh it has to do with the fact that this three D uh plus the and the fast
0:12:30	so that was
0:12:31	really thought telling you about the convolution know as it i mean as a a a uh than men's the
0:12:38	i mean if they can be
0:12:40	that's a can be then because um P T
0:12:43	satisfactory this factor
0:12:45	to to have a a a a a a lot of the challenge is and you have the set of
0:12:50	image analysis because
0:12:52	this is a a and the by just to have numbers
0:12:55	and the so the a set so the to uh uh my west image shows which have to do with
0:13:02	because because tracking so this self worse more those in images
0:13:06	looking that's that was also a
0:13:08	i have the show if the gene expression
0:13:11	for a has and and you know some the some of the problem of that that to the
0:13:15	a a a a a and to for for image processing
0:13:18	so i
0:13:20	i i i i will actually this is the sort of give you some example state of the art
0:13:24	and taking a a uh algorithms that have been developed a people in a common T that she i been
0:13:31	used by religious
0:13:32	and i mean this is this that of the to the the challenges that i to go on the and
0:13:37	i with
0:13:38	sure but
0:13:39	for example here
0:13:41	okay so if if
0:13:42	this
0:13:42	you see here you have
0:13:44	this is an example of a uh uh you know time laughs my first P
0:13:49	where uh uh actually this is a a ms so so
0:13:52	uh uh uh the end of the proposed um that has been
0:13:55	the the the uh is inside the nucleus and have
0:13:58	the are just like to
0:14:00	to is month so for of those from more
0:14:03	and so that was a better here is the tracking i
0:14:07	uh that will actually five
0:14:09	true
0:14:10	in so i'm and and i is to do that
0:14:13	if are less fashion using
0:14:15	oh to defined as function using the fast
0:14:20	and
0:14:20	i mean this is
0:14:21	uh just works very well for
0:14:23	a single party
0:14:24	but the
0:14:25	i'll the challenge
0:14:27	is uh
0:14:28	you may have this type of
0:14:30	okay
0:14:31	so you have
0:14:32	yeah yeah of those party
0:14:35	oh have of for and the background
0:14:37	and so uh a real challenge
0:14:40	design
0:14:41	for for for for that
0:14:42	data so it's
0:14:43	you know
0:14:44	you'll three
0:14:47	a uh so next challenge
0:14:50	uh so it has to do with
0:14:52	self
0:14:52	uh and uh uh so uh shape and
0:14:55	you need it
0:14:55	so he it is simple
0:14:57	cells
0:14:58	E
0:14:59	so the violent adjust here this is the uh
0:15:02	face got five
0:15:04	my cross is so
0:15:05	yeah i i is to outline of cells
0:15:08	not just for counting them but because we want to extract the gene expression profile
0:15:13	it's so this is also like can of it
0:15:15	the ball of that's
0:15:16	by
0:15:17	by all
0:15:18	will
0:15:18	kind of like
0:15:20	hmmm
0:15:20	a fact let's
0:15:21	images
0:15:22	and then if you uh uh
0:15:25	like that of
0:15:28	oh
0:15:29	you're
0:15:30	so that what what you have
0:15:31	principle here
0:15:33	the
0:15:35	so what's important here it's
0:15:37	for all those cells
0:15:39	with time so why would you want to do that
0:15:41	because all cells actually what
0:15:43	right thing
0:15:45	a different genes your
0:15:46	that'd been through recently label and and so here you you you know the outline of this
0:15:51	the the cell
0:15:52	but what you would like to quantify is the expression
0:15:55	oh uh i mean the amount of fluorescence in in in both cells here
0:15:59	and and so get that uh time course of of fluorescence and and and and and uh this this will
0:16:05	be the date so that the biologist one two
0:16:08	you we uh i mean to extract so what a
0:16:11	here uh the challenges i mean the challenges is very poor contrast of those images
0:16:16	dealing with much more complicated it's uh shapes than those used cells which are round
0:16:21	and things cell division it should using models of time evolution
0:16:25	uh and doing global optimization in space and time
0:16:29	dealing with crowded images touching cells in to reducing repelling forces high throughput puts mean this needs to be done
0:16:36	on you amounts of cells and power that imaging
0:16:39	fast row put able and the uh above all easy to use a gram so that people uh i really
0:16:45	applying the mean brock
0:16:46	so now uh relating more to what to a lot was telling you
0:16:50	uh uh you know there this a problem of extracting features from images and and so for example this was
0:16:57	also
0:16:58	a program developed
0:16:59	it's the
0:17:00	it's a program for tracing uh acts on of of neurons but
0:17:04	at some your to me
0:17:06	and and not
0:17:07	i mean i mean this works well it's uh use by by people
0:17:10	but uh uh uh
0:17:12	i of course uh i mean life is is more complicated than that
0:17:16	and what i see as a challenge an opportunity uh a of uh for for for a signal processing
0:17:22	community actually coming up with
0:17:24	uh what that would call key point for by you because key points have the huge six S for computer
0:17:30	vision so is looking like
0:17:31	points of interest in image so what we would need a by all we are not like we have three
0:17:36	D data
0:17:37	so we would not need for well we can first look at two D but optimize
0:17:43	a for scale uh a translation rotation invariance
0:17:46	also i mean what we're seeing we have background verses uh a structures of interest so doing detector that we
0:17:53	might seem i
0:17:54	uh suppress press background and
0:17:55	probably use this kind of idea for designing wavelet type representation that can
0:18:00	and has the feature of interest
0:18:02	and and and and suppress the background
0:18:05	but the real problems in three D
0:18:07	okay because there's nothing in three
0:18:09	oh O case to two minutes okay
0:18:12	and and and and so i that she do that in three you
0:18:15	so in three you have a interesting structure like sheets membranes
0:18:19	and he the challenges developing that's a durable the vectors wavelet since really
0:18:24	and uh above all be computationally efficient uh uh because i mean the are huge amounts of data
0:18:30	so i just wanted to
0:18:32	uh here are uh i i mean this is perhaps what where banning in the lab but just to show
0:18:36	you that
0:18:37	uh one can do a very interesting steerable wavelets so we here are
0:18:41	i'm in in in in two D that look like has since but there a reversible
0:18:46	uh where let's but the
0:18:47	this seems also to be a possibility to do them in three D and this is uh
0:18:52	pretty much uncharted so doing
0:18:55	where it's that can what's agents
0:18:57	and i yeah side like just like to have a size
0:19:00	if you're working in biology so that
0:19:03	it's a disciplinary research where there are lots of players of course the to just the medical people
0:19:09	there's the optics the microscopy the by chemistry which is the mark and the signal processing
0:19:15	and there were there was also like here
0:19:17	special issue on biological imaging a few years ago but is still a good entry point for those want to
0:19:22	get in uh in with few okay case
0:19:26	of
0:19:27	X
0:19:57	yeah
0:19:59	as so uh i will
0:20:00	and the law speaker go a bit more into challenges that are you merging uh especially in the medical field
0:20:06	and
0:20:07	moving
0:20:07	wait a little bit from the biological
0:20:10	um so basically number one chance number one is still increasing data that now
0:20:16	already already for treat this is going on
0:20:18	uh the the simplest by medical signal is the
0:20:22	electrocardiogram the
0:20:24	second two dimensional signal is the the
0:20:27	normal two dimensional picture
0:20:29	then uh in the eighties came to three dimensional pictures
0:20:33	and then the four dimensional pictures
0:20:35	being
0:20:35	a three dimensional close
0:20:37	time
0:20:38	uh more recently especially in the the last decades uh specially functional imaging
0:20:43	on top of and apple
0:20:45	imaging has become very cool
0:20:46	in that case you three spatial dimensions one time dimension
0:20:49	one
0:20:50	functional entity for instance uh blood oxygen nation
0:20:54	um
0:20:55	three plus three dimensions basically meeting three
0:20:58	a a spatial dimensions a three direction vectors uh base
0:21:02	uh
0:21:02	diffusion tensor imaging
0:21:05	then there's uh nowadays very popular uh it's starting to really get popular
0:21:09	three plus three plus one
0:21:11	and dynamic uh imaging
0:21:13	for
0:21:14	where you can actually a monitor or you know that the
0:21:17	a three dimensional flow
0:21:19	inside the ha
0:21:20	and especially signal processing for
0:21:22	this is still in its infancy this sense that
0:21:25	able try to characterise a for text flow right
0:21:27	he's uh in
0:21:28	he's types of images
0:21:29	but then when you have
0:21:31	a i thought we kind of reached a limit at uh seven dimensions here
0:21:35	but then uh this uh
0:21:37	and and brain database was published
0:21:40	and a brain database is basically three dimensional data dataset
0:21:43	plus at time aspects where it basically for
0:21:46	a my
0:21:46	of mice
0:21:47	where for every voxel they have twenty thousand G
0:21:50	which uh
0:21:52	values for gene expression show basically
0:21:54	but see this as it
0:21:55	you which uh uh dimensional data set
0:21:58	a a which is now uh just about to be explored
0:22:01	to correlate it's uh with all kinds of
0:22:04	for imaging
0:22:06	so this is number one the dimensionality of the data is still increasing
0:22:10	rapidly
0:22:11	and signal processing challenge
0:22:13	are
0:22:14	the uh plentiful there
0:22:16	seconds is that um
0:22:18	imaging
0:22:19	uh until about ten years ago was mostly about imaging structure and it to me with ct anymore
0:22:25	yes
0:22:26	and function
0:22:27	with for instance uh a nuclear technique
0:22:30	it would highlight some aspect function
0:22:32	and past decade all kinds of technique became available to also look at biochemistry
0:22:38	and
0:22:39	so now we can look at diseases on a biochemical level
0:22:42	and then see what happens first in biochemistry uh what goes wrong in a particular
0:22:47	disease
0:22:47	by chemically
0:22:48	and then study effects all structure
0:22:51	and function
0:22:52	now the data acquisition is now
0:22:54	getting so far that we can acquire these triangle structure
0:22:57	function and by chemistry
0:22:59	but
0:23:00	the signal processing and image processing is still
0:23:03	largely to we think this on as separate entities and
0:23:06	i i think there still a big challenge
0:23:08	in trying to come up with in to great if
0:23:10	signal processing
0:23:12	it actually looks at all data at the same time
0:23:16	there at very important trends uh i i think shows also that the
0:23:20	few of by a middle image and that was is is you ring
0:23:23	at at this point in time
0:23:25	used to be the case that you could get away with a validation of a segmentation algorithm for instance
0:23:30	just by citing some uh uh uh accuracies that others have achieved
0:23:35	and then uh a benchmark your algorithm
0:23:37	to that but the problem with that is that
0:23:39	all those those uh
0:23:41	results published in these papers
0:23:43	they are usually done different patient different test state that different test protocols different error metrics different gold standard everything
0:23:50	is if
0:23:51	what you see and this make it very difficult to really make an object is mark
0:23:56	and especially in the pattern recognition community there's already a lot of standardised databases we
0:24:01	known classification result
0:24:03	we see now
0:24:04	in the medical image processing field
0:24:06	more and more
0:24:07	trends
0:24:07	words we objective quantification they make fable
0:24:11	a centralized data
0:24:13	evaluation scripts are run sense really
0:24:15	and everybody can benchmark their data their algorithm against that particular that's it
0:24:21	i i think this trendy so input
0:24:23	and that
0:24:23	nowadays if you try to publish uh and algorithm on
0:24:28	a topic which has already been that smart
0:24:30	in one of these
0:24:31	challenge
0:24:32	a you cannot get away without
0:24:34	uh benchmarking marking it against it
0:24:36	and there's kind of a default rejection
0:24:38	if you do not include validation experiments on these standardised uh um
0:24:43	data bases
0:24:44	so very important field of of of yeah it shows that the
0:24:48	the the the field is but your in a that's
0:24:52	for if very important new topic uh is a key to more data analysis
0:24:57	right the main question to answer is
0:24:59	something is changing but what
0:25:02	now and you can imagine that
0:25:04	if you look at an in fit individual patient uh yeah it still manageable
0:25:08	uh you look at follow up
0:25:09	uh is
0:25:10	easy is getting worse yes or no
0:25:12	uh but nowadays one tends to look at
0:25:15	groups of patients anyway between a two
0:25:18	and one thousand
0:25:20	that
0:25:20	this study D V disease each development in two weeks writes treatment affect
0:25:24	but
0:25:25	even more
0:25:26	data is acquired
0:25:27	normal subjects there and now you huge
0:25:30	population studies right there
0:25:32	every year scanning people healthy people
0:25:35	and they're is getting every here and then basically wait until people get sick
0:25:39	at some point in time
0:25:40	and then they can backtrack
0:25:42	through the date that whether they can find
0:25:44	early signs of that
0:25:45	you can imagine that that finding these are signs is really a needle in a a a a stick
0:25:50	and uh well there's a huge uh image processing challenges
0:25:54	trying to really combine all that data and to backtrack to bit data
0:25:59	um
0:26:00	well another trend to actually here is uh
0:26:03	a fact that as that already showed that that
0:26:05	these biochemistry a chemistry became available
0:26:08	uh i think there is a there's a lot to gain in the combination
0:26:12	of chemistry tree
0:26:13	and
0:26:14	specified as signal sensor specific signal sensors
0:26:17	that for can look at multi spectral
0:26:19	imaging as we can see here
0:26:21	and and let's say
0:26:23	one of the applications that we are working on a second
0:26:25	probe
0:26:26	a specific
0:26:27	for a probe which is injected for four it to more
0:26:30	a and that uh basically that broke can be used to identify
0:26:34	this the lymphatic uh system and then
0:26:37	it's very easy
0:26:39	for
0:26:39	and some parts of
0:26:41	particular search you really need to remove lymph nodes
0:26:43	and this is only
0:26:45	possible because of a very delicate balance between and so point
0:26:49	acquisition device and
0:26:51	the probes that are being used in here you can actually see
0:26:53	that's removing the lymph node becomes very easy
0:26:56	using these combines a
0:26:59	in in finally
0:27:00	uh especially uh uh yeah recently there there is immersed a lot of
0:27:06	attention for personalised integrative modelling modelling and simulation
0:27:11	um
0:27:12	perhaps some of you have heard of the virtual physiological human
0:27:15	project
0:27:16	well especially you this has been a whole public for the past few years
0:27:20	and more and more this to these uh very
0:27:24	integrated models
0:27:25	go from the cell like organism
0:27:27	they start to percolate we into clinical practise
0:27:31	uh yeah we're actually they use this this uh uh
0:27:34	uh
0:27:34	patient specific data
0:27:36	tuned to mobile
0:27:37	to the observations and from there um they can actually do predictions and simulations
0:27:43	so that
0:27:44	bout wraps up my uh my uh overview of uh
0:27:48	what i perceive as very important
0:27:50	to trends in
0:28:40	i think we need them both
0:28:42	basic
0:28:43	uh i think of some
0:28:45	i L
0:28:46	some algorithms uh are so generic that they can basically
0:28:50	for for
0:28:51	for a lot but i think
0:28:52	we also need dedicated
0:28:53	times algorithms of four
0:28:55	particular application
0:28:56	perhaps you
0:28:57	i actually my my advice there is uh for you know for those are not been working too much uh
0:29:03	in
0:29:03	in the body oh area
0:29:04	my advice what would be a good start working
0:29:07	on the specific algorithm
0:29:09	because
0:29:10	you look at the very big challenge you know the ultimate segmentation algorithm you'll have to compete against the
0:29:16	all those benchmarks that that are out there so if you
0:29:19	if you're working on the specific yeah a problem so you'll have people very happy at the other end
0:29:25	and and and and so it will really help you also to learn
0:29:29	what what the issues and of course i mean if along the way you happen to stumble on something really
0:29:35	general
0:29:36	well then
0:29:37	you go also for for for the more general albert but i think specific is
0:29:41	good
0:29:45	there has been
0:29:45	it's went let's say in the nineties to look for this uh big general super segmentation framework for instance that
0:29:51	could be applied just to everything
0:29:53	and now
0:29:54	i see kind of more dedication again to single applications
0:29:58	so that
0:29:59	and and only some of those very generic
0:30:01	uh segmentation framework
0:30:03	yeah
0:30:04	they're they're still uh
0:30:06	actively researched
0:30:09	yeah i mean
0:30:10	yeah
0:30:10	in most of
0:30:11	areas like segmentation i mean the generated stuff is already in that
0:30:16	i
0:30:17	well maybe not all music you're very lucky
0:30:20	you'll find
0:30:21	your generic i'll
0:30:31	uh maybe uh yeah since we don't have many many uh questions
0:30:35	uh so we
0:30:37	i guess in our presentation we did it's just too much on the inverse problems because we thought every someone
0:30:43	else would talk about
0:30:45	"'cause" the you will compress and you or uh
0:30:48	a those L one type of to
0:30:50	as a a pretty much in the focus of of the signal processing community
0:30:55	and and the a search this stuff with say
0:30:58	actually those a a a a a a very hot topic in in in imaging in general so
0:31:04	essentially essentially people are read is it in all the classical uh
0:31:08	uh uh image reconstruction algorithms the you for "'em" C T
0:31:13	uh
0:31:14	in in the medical imaging area a
0:31:17	the same uh for for for for different
0:31:20	a could be source so that's a
0:31:22	and and and not just the the traditional
0:31:25	uh but that each is that also like lots of people and optics so uh designing all always new new
0:31:31	a new modalities and
0:31:32	then you mode that it is uh ten to work very closely with signal processing because
0:31:37	traditionally for example be plenoptic optics they would just want to to
0:31:41	to to do a microscope so that you see an image
0:31:44	but now uh uh once you have signal processing you can do lot more like a tomography
0:31:49	start a acquire measurements that you may not necessarily see but once
0:31:54	the goes through a signal processing group
0:31:57	then that the can can we construct images hopefully using that's measurements
0:32:02	and so forth and so this all obviously they hot so if you go to medical imaging
0:32:06	call you have lots of compressed sensing people the same on on the microscope beside
0:32:12	so so that's so
0:32:13	uh a there are yeah where the problems up pretty you well be fine
0:32:17	but but then
0:32:18	uh the guys also a very interested in a quantitative
0:32:21	measure
0:32:22	so
0:32:23	suisse showed in our example
0:32:26	just to
0:32:27	just to elaborate on a michael was saying is that
0:32:30	but but for than them
0:32:31	by your logic logical um can
0:32:34	the goal has always been to do a nice image and nice image
0:32:39	as you may know is
0:32:40	probably don't the base one in terms of
0:32:42	uh optimising devising the the rate of acquisition the rate of uh information
0:32:48	it's very but the opposite
0:32:50	i i actually
0:32:51	if you
0:32:52	a through signal processing are able to extract the meeting for information even if there is a lot of bad
0:32:57	run or clutter
0:32:59	to means that you can
0:33:00	either improve
0:33:02	the
0:33:03	the timing of your acquisition you can expose more time you're your sample
0:33:07	or you can
0:33:08	uh also spend more time on doing um different ranges of uh wavelength
0:33:13	so you can
0:33:14	have a more dense
0:33:15	a field of acquisition if we were able to reduce the time that you exposing example for each of the
0:33:21	the top
0:33:22	in this is something that only signal processing related
0:33:25	method can bring to the community so there is a whole
0:33:29	range of application enough
0:33:31	possible uh
0:33:32	topics of priesthood for these comedians icassp
0:33:35	if only you are interested to
0:33:37	i mean uh the other with uh some challenging but also interesting problem
0:33:45	not to mention the con the also the
0:33:48	the fact that we are more immoral maybe use of the plantation there is some kind of a coming together
0:33:53	of uh
0:33:54	medical imaging in biological imaging and they are really merging
0:33:58	in one of the key uh
0:34:01	but all that is is molecular imaging which is in the the use of medical imaging techniques
0:34:06	to have
0:34:07	resolution which is similar to uh microscopy uh imaging
0:34:11	in there there's also a lot of challenges still opened for the crib
0:34:19	well there are no more question i think needs O
0:34:46	uh
0:34:47	i i just saw on the medical imaging side uh maybe
0:34:52	not everyone will agree with me but uh it's pretty much your because uh i mean the big revolution happened
0:34:58	about thirty years ago
0:35:00	uh invention of summarise so i mean M R I still going very very strong and always getting better
0:35:07	stronger magnets et cetera
0:35:09	uh a new type of modality is drink different kind of measurements but
0:35:14	it's pretty sophisticated
0:35:16	now what's happening on the other hand on the biological side
0:35:20	we are
0:35:21	more or less experiencing this
0:35:23	state of thirty years ago in medical imaging because they're all kinds of new modalities actually coming out
0:35:29	almost every year
0:35:31	and for example in microscopy they've been able to beat the the relay team it's by a factor of one
0:35:37	hundred going
0:35:38	you know like
0:35:39	below what physics still
0:35:42	but but by using some tricks of course and
0:35:44	and they're like a novel microscopy is being developed the current the and very often
0:35:50	hand in hand with with signal processing so so that's very important actually the the other thing we could even
0:35:56	say it's all the medical imaging couldn't exist without
0:35:59	signal processing because a mirror i the first thing it is it's the fourier transform
0:36:05	and and and the
0:36:06	so uh so i see that uh there there i mean there's improvement in in modality especially
0:36:13	uh going out to find the resolution
0:36:15	uh and and very much happening now in in the area of biology but i suppose also
0:36:21	it equal so i don't know if few and this all of this small animal imaging
0:36:25	yeah or uh uh so so where we getting always uh need for higher resolution source so it's very very
0:36:31	act
0:37:01	oh i i i mean this is very very hard because of for example with my cross could be a
0:37:05	good you use of your essence and there's lots of
0:37:08	uh a uh you know like naturally fluorescent so the
0:37:12	stuff
0:37:13	inside you used specimen so usually
0:37:15	i i mean i mean this uh
0:37:17	this will be a uh uh you will interact with some biologists to do the the lot labelling and and
0:37:23	and i mean it's never a very simple
0:37:25	but of course i mean the goal a goal of biochemistry is is finding mark is that they extremely selective
0:37:31	and and also with this molecule image in principle i mean they're very good marketers soak is just
0:37:38	sometimes the
0:37:39	the the resolution is
0:37:40	it's so great yeah so uh so we like with but um they very good mark but very poor was
0:37:45	should
0:37:45	with a by you them in this since very good mark "'cause" but that terrible resolution
0:37:50	and and
0:37:54	K i think we have to close the session the "'cause" the next uh station is a starting and uh
0:37:58	i think you all for your or of coming here and uh hopefully will see you in some more dedicated
0:38:03	uh
0:38:04	conference is of stations to uh imaging thank you

Trends in Bio Imaging and Signal Processing

Expert Sessions

Presented by: Jean-Christophe Olivo-Marin, Michael Unser, Laure Blanc-Feraud, Andrew Laine, Boudewijn Lelieveldt, Author(s): Jean-Christophe Olivo-Marin, Michael Unser, Laure Blanc-Feraud, Andrew Laine, Boudewijn Lelieveldt