Přepis řeči - SIGNAL PROCESSING IN MEDICAL IMAGING AND IMAGE-GUIDED INTERVENTION

0:00:13	i think is to come
0:00:14	"'cause" to come here if you can still hear me from here
0:00:17	a a small enough
0:00:18	room
0:00:19	should be easy
0:00:21	so
0:00:23	a medical imaging an image analysis uh
0:00:26	a really is uh
0:00:28	it a topic which is so
0:00:29	very widely used in the routine clinical practise these days and they especially image acquisition part has a revolutionised the
0:00:35	might is in a uh today
0:00:37	uh
0:00:38	a probably out of a of that the X rays are being with us from more than a one hundred
0:00:42	years and that
0:00:43	uh
0:00:44	first of accidental a medical image of of of missus run again
0:00:47	uh uh was not in a ninety five
0:00:50	and uh after that the right of different the a you mean you wanna only this sorry there exist that
0:00:55	but is uh i no x-ray ct was magnetic resonance imaging with a was that are emission tomography
0:01:00	with a single photon shouldn't
0:01:02	no tomography with optical coherence tomography
0:01:05	and that out of similar
0:01:07	more these which are used to really widely and uh
0:01:10	and they them many ways uh have changed the way how physicians do the madison
0:01:15	and that many ways a you know the physicians are virtually
0:01:18	unable these days uh do uh function as the used to
0:01:21	uh even ten or fifteen years ago
0:01:25	the an existence or medical imaging cut is really heavily depend then on the advances in the signal processing and
0:01:32	that that's why V be invited to a present here
0:01:35	and to to string some the ties between the signal processing community in the medical imaging community
0:01:40	would the ability uh yeah to image in people biological structures is uh uh uh true dependent on the signal
0:01:46	processing the the signal processing is making a
0:01:49	the the in a blink uh engines uh you know behind the scenes to get the images uh i in
0:01:54	two D three D four D and and a five D routinely these day
0:01:58	so
0:01:59	the three D image is probably obvious or at is it's a three-dimensional volumetric image which we get
0:02:04	the for D maybe a little a little tricky
0:02:06	so if you imagine you know a a beating heart uh that's a three dimensional object which are has the
0:02:11	force the this time
0:02:13	and uh now you maybe even trickier able to five B may mean
0:02:16	so imagine the beating heart uh which is a four dimensional uh and to the
0:02:20	and then you do the imaging over time for example every week on every day and you could the films
0:02:25	the you know the imaging
0:02:26	and that that's probably uh
0:02:28	the suspect to dimensionality
0:02:30	uh how far you typically get older you can think about that combining multiple of D imaging can images and
0:02:36	the and the go to the higher D's uh you know and that the that matter
0:02:40	there are two uh
0:02:42	main ways how you can i didn't you can acquired the images
0:02:45	uh
0:02:46	maybe one of those uh
0:02:47	one possible distinguishing fact would be very are rising ionising or non housing addition
0:02:53	X ray being the ionising radiation of course uh for example E um
0:02:57	you know i'll just some big not ionising
0:02:59	still
0:03:00	or you may have passive if listening to signals like a
0:03:03	they don't so i think a your you're more like you was uh uh in some ways and listening to
0:03:07	do the results are like in M mr
0:03:10	let's talk a little bit of bottles individual imaging modalities uh
0:03:13	and uh
0:03:15	you know one of them maybe maybe do to
0:03:18	one a very exciting ones of course is the x-ray ct which is very routinely used for three D and
0:03:23	four D uh imaging
0:03:25	and that the image a
0:03:27	reconstruction the the
0:03:29	formation of the image is heavily depend then on the radon transform and and similar transform which you can do
0:03:35	oh so the image reconstruction is not from projections and uh it's uh
0:03:39	oh obtain from their you think x-ray beam
0:03:41	there are many advanced the reconstruction techniques and of you hear about the some of them today
0:03:46	uh from a michael or and and you rang
0:03:49	uh including the multi beam scenarios and than found beam scenarios and someone
0:03:55	the recent recent uh x-ray imaging is combining multiple X source is a multiple x-ray detectors
0:04:01	and that
0:04:02	those uh source is and the text as our day thing in the uh a very high speed around the
0:04:06	body was about point to eight
0:04:09	i think is the highest the current speed the available in the in the commercial C Ds as the extra
0:04:14	scanning speed actually exceeds forty centimetres per second
0:04:17	which means that you can uh image to act chest the
0:04:20	uh a region of a human body in less than half a second to you can't the entire body
0:04:25	uh image about five seconds which clearly is causing potential problems with the patient leaving the table if you stop
0:04:30	the scanner and the
0:04:31	no
0:04:32	just starts getting because the speed of the of the uh you know a card uses so so far
0:04:39	so it one example i also try to show some examples of the individual limit what of the images
0:04:43	this is an example of a roughly a domino uh part of the human body you can see the the
0:04:48	spine in the middle
0:04:49	um you
0:04:53	you may be able to see some of the uh
0:04:57	some of the vessels uh which you can have here
0:04:59	and that
0:05:00	uh
0:05:01	you can identify the inner and outer uh walls of the vessels well for example if you look at this
0:05:05	particle i we can see that
0:05:07	but being to white part of this
0:05:09	this uh a green area being the to wall and really we can difference between this particle slice and that's
0:05:14	lies it seems to have a normal wall
0:05:16	seems have some additional material
0:05:18	uh inside of uh
0:05:20	of the room in a wall
0:05:21	which are if you image in three D will find out is somewhat from baltic area
0:05:25	uh in this uh in this location
0:05:27	and that a something which should not be there
0:05:29	from in your is more changes it can be
0:05:31	you by X ray D in this case a some contrast that will being included in the
0:05:36	in the book
0:05:38	oh example from the x-ray ct
0:05:40	these the example from the long imaging
0:05:43	where we would have one slice or of the make the
0:05:46	uh
0:05:46	uh mid level of the long
0:05:48	uh you can see the uh
0:05:50	the left and right long
0:05:52	you can see the individual features here like this portion here this portion there which uh divide quite nicely quite
0:05:57	visibly
0:05:58	the individual so
0:05:59	of uh of the long
0:06:01	and of course you can get a sweet michael are presentation of lungs with the colour coded the all
0:06:06	right to identified
0:06:07	a back to that little bit uh
0:06:10	if you talk about E mr imaging again it's three D forty the imaging modality
0:06:14	which uh
0:06:15	uh start with the high strings made tick feel that lines the minimisation of the hydrogen atoms water
0:06:21	and uh
0:06:22	you of
0:06:24	boy some a radio frequency also
0:06:26	which would core use form the whole dating the tick that and that this of can be detected by D
0:06:31	M are scatter
0:06:32	and then us
0:06:33	and that
0:06:35	you create we need a sense a different all some very complex signal manipulation to get that was in view
0:06:40	images
0:06:40	but the images are quite spectacular
0:06:42	so in this case a you can see the example of the cardiac and is is the beating heart in
0:06:47	the different projection to a short X is long X image
0:06:50	uh uh of the left and the right ventricle so left and right but suppose here or you can also
0:06:55	go uh so was a slice of like a short axis lies which should go no from
0:06:59	the valve of all the way down to the a set and that you can see how it looks a
0:07:04	uh a three do might for the and the asked like a face or a and cyst like face
0:07:08	it would
0:07:09	the like station and the screen is uh of the hard to
0:07:13	it's a lot a a way how you can image do the same area of the body
0:07:17	and that you have the hard which we so before like here but you have a different the direction of
0:07:21	uh of your slices
0:07:23	and this case on the left and side you can see D left ventricle out fault tracked with the aortic
0:07:28	valve
0:07:29	in this part you see what is called a candy cane uh
0:07:31	the christmas you can to came uh you
0:07:34	where a you see the or that quite nice still do i way out
0:07:38	uh to the that
0:07:40	if an example of mr imaging curve from the area of car a it's a from the neck area
0:07:46	where in the uh
0:07:48	bought for uh
0:07:49	uh
0:07:50	vascular a image you can see that this ten is the narrow incur of the vessel right here
0:07:55	and that uh you can look at in the view cross section so off
0:07:58	uh
0:07:59	all the vessels
0:08:00	if you look for example at the two different cross sections like a in this yellow and an orange areas
0:08:05	you will see that that there is that that car date that
0:08:08	uh
0:08:09	right here with a little bit thicker wall
0:08:11	and that that's
0:08:12	well a good thing but it's not that super bad thing but if you see on the on this difference
0:08:17	lies
0:08:17	a slide that you sure by orange
0:08:19	that that the room and start be really late i
0:08:22	that and that you are getting very close to having a true problem
0:08:25	and that this uh a find would be a source to strong uh
0:08:28	which which clearly is not a
0:08:30	a joyful think that have
0:08:32	uh are you know don't have to image or only the cardiovascular system
0:08:36	is another example of the knee joint the imaging
0:08:38	what you have the the femur the
0:08:41	a T V and about that one here you can see the bone structures as well as the individual portions
0:08:45	of the cartridges
0:08:46	which uh
0:08:47	uh which you have you can analyse and the uh again shows sold
0:08:50	just like
0:08:52	if you switch to the ultrasound the
0:08:54	and the old D optical coherence tomography images again we can easily do three D and forty uh imaging
0:08:59	the ultrasound is using ca
0:09:01	uh
0:09:03	of of sounds uh
0:09:04	of a different frequencies frequency the frequency influence the resolution of the imaging as well the depth of penetration of
0:09:09	the signal
0:09:10	but a money whatever T
0:09:12	i know that you can again use it for
0:09:14	a variety of uh best Q and the
0:09:16	you'll be jane uh
0:09:18	uh i don't techniques and so
0:09:19	you have to go coding the model or face in some way similar to the ultrasound
0:09:23	uh uh it uses different signal processing cut approach is with the coherent light
0:09:27	uh it allows uh
0:09:29	to a
0:09:30	choir images of very high resolution
0:09:32	uh based on the uh frequency laser
0:09:35	uh maybe a hundred and a a one micron
0:09:38	and that
0:09:39	therefore lapse of penetration of we'll all but it's of it's fess found uh
0:09:43	very fast a fascinating applications of non a six three D imaging
0:09:47	view of the red in a and the core black cat
0:09:50	uh
0:09:51	from the inside a a interest vascular image
0:09:54	i the example the ultrasound sound some a two D has slice is uh you can see the individual of
0:09:59	use uh you know the heart
0:10:01	you can also do a imaging what is called a real three D echo we four dimensional model the
0:10:07	where a the and entire three D volume is acquired at the at the same time of the real time
0:10:12	we get for example compared to D which is a don't not perfect that is respect to noise with a
0:10:17	three D which is even rest
0:10:19	picked to because you have uh
0:10:20	a little bit less time to do all the audio
0:10:24	an example is uh coming from but
0:10:26	uh in geography imaging interest lots of sound imaging
0:10:29	a that on the on the left and side that uh
0:10:32	i can start okay level inside you had the x-ray projection of image
0:10:36	with the not the K area here you can in they call that it with the intervals ultrasound sound that
0:10:41	for on the tape which all dates that that thirty frames per second
0:10:45	you can get the imaging if you put back the cost are
0:10:47	uh use of can get a dimensional image or a of the current are three you see the vascular structure
0:10:52	to the the any here this is the or itself
0:10:54	as well as the
0:10:56	the black which is uh
0:10:57	a which may be uh it all okay the in in the corner
0:11:01	this is an image or from the retinal channel uh all C
0:11:04	uh uh image or of the macula
0:11:06	oh so for the visions body use the for we have it's is that in the nation the
0:11:10	uh a location but really see sharply this C D optic nerve had the blind spot though of the right
0:11:15	a
0:11:16	it's very interesting to identify those individual layers and start to associate the think is of those in the you
0:11:21	or so with a
0:11:22	um individual is easy which at which
0:11:26	so the signal processing in image acquisition is uh use the very broadly and that
0:11:30	uh clearly the advance is uh and the complexity of signal processing pipelines are
0:11:35	absolutely critical for uh medical imaging medical image position
0:11:40	we are trying to get heist image quality a high speed of acquisition highest resolution that's all what signal processing
0:11:45	is gay
0:11:47	uh a a question which is here is a how to get the high school of the images is minimal
0:11:51	goals of and we are working with ionising radiation again a challenge the signal processing community
0:11:56	and that something what the what be
0:11:58	have need the
0:11:59	uh to be able to use uh the imaging
0:12:02	the best
0:12:03	oh i made a statement and the very beginning that that
0:12:06	uh image acquisition has revolutionised the critical uh portions of the madison
0:12:10	the image analysis has not
0:12:13	and that image analysis is a a cold in medical image analysis is widely used as research applications
0:12:18	lights
0:12:19	not
0:12:20	for fully enter the clinical as
0:12:22	it's to in clinical except and
0:12:24	and that
0:12:25	yeah
0:12:26	robert but a can i it's of corn there is all the first one it really would you use the
0:12:31	in the in the quantity matter
0:12:33	uh it's also used for a screening in the cardiovascular matter in of tonic uh imaging matter and so on
0:12:38	so for example if you can could occur it into a media thickness measurement is it the car the are
0:12:43	three now image the
0:12:44	uh
0:12:45	with the ultrasound you to to stick this of the wall is the in a bright here the media layer
0:12:50	you in black
0:12:51	and that i can start analysing the thickness of the wall
0:12:54	and that you can uh measure
0:12:56	what think is it and the this case the thing it's of the new a wall this one would be
0:13:00	but i have a many be a point six million you here
0:13:03	you may have a different a subject that body measure to think it's of the wall to find out of
0:13:06	stick this is actually about the many meter for the farm
0:13:09	now the question is what i me
0:13:11	and uh you can look at norm at if uh a a so of normative data is a a uh
0:13:16	uh of uh of patient is respect to a H you may find out of for example complete normal
0:13:22	while the second example is actually think wall and
0:13:24	some person it'll will be a high risk a
0:13:26	of uh
0:13:27	but a heart attack or stroke a
0:13:29	uh sometimes later in life
0:13:31	and the uh maybe be ready for
0:13:33	uh uh or some
0:13:34	a drug treatment that
0:13:35	uh a in something
0:13:38	i showed do some of the uh images image is all the or to just to give you an idea
0:13:43	how the analysis looks like uh you can get a three the actually four dimensional a analysis a three D
0:13:48	plus time of the entire aortic blanks this respect to cross sections and motion and again gain compared to a
0:13:54	a normal values maybe some values of uh
0:13:57	um
0:13:58	of uh subject to
0:14:00	with some E
0:14:02	real three D echo he's C D image back here i K i all that you can see those also
0:14:06	those red calm to was on the right hand side
0:14:08	in automated analysis which a also you to get maybe to uh volumetric metric information
0:14:13	we the cardiac cycle partly ejection fraction what meant not on the base on the area
0:14:19	and a area from a uh x-ray ray ct
0:14:21	uh from the image of the long
0:14:23	you can get the entire at every rate be analyzed you can a
0:14:26	use it for uh
0:14:29	for
0:14:30	of image guided the surgical makes you want to reach some location of a you know how to get there
0:14:34	you can uh do to separate the noise is one do you all and and you segments of the long
0:14:39	uh
0:14:40	a you probably here
0:14:41	bit more
0:14:42	about that
0:14:43	uh
0:14:44	we do have uh i don't a we have for context stress analysis uh uh for the for the me
0:14:48	again image show showed you before
0:14:50	where we identify individual balls identify the cartilage locations
0:14:54	and that
0:14:55	based on that you can cochlea in view by mechanical stress so which are applied to
0:15:00	a individual subjects a
0:15:01	uh being included in that uh imaging
0:15:04	which
0:15:05	oh of course you don't image only hear men's we image uh animals as well including small animals
0:15:10	so this is an example of a models this is the most long for the mike C you scatter
0:15:14	and the uh
0:15:15	a reconstructed that three dimensional uh at a rate the of the mouse
0:15:19	which uh uh seems to be an important thing
0:15:22	or example from the malls uh with a three D C give registration many or image the same also over
0:15:27	time you to register the models uh images
0:15:29	uh so that you can find out where exactly the same point was in different scans or you like to
0:15:35	it'll get a made use a like a later uh
0:15:38	uh
0:15:38	for different my
0:15:39	this is the example of the analysis off
0:15:42	that the macula make or image at uh from all C T in you can see we identify a large
0:15:47	number of individual layers which are uh are are able to identify uh
0:15:51	get the link use with uh uh are be is for example and some other uh disease
0:15:57	i have one example whole for visualization a guide intervention here
0:16:00	uh from the uh
0:16:03	from the a uh
0:16:05	surgical planning so traditionally you are planning on the on the two D display
0:16:09	and uh
0:16:10	you can also do the surgical i think in the virtual reality and wider
0:16:13	and that you like to have like the three dimensional view
0:16:27	and this is a a short movie that swap shows uh uh what would be interesting visualisations which are you
0:16:33	have down to lie about together with the across university of technology incoming still grad
0:16:38	you see the labour uh being ca three D visualised with the
0:16:42	uh vascular trees and the tumour does the
0:16:44	as the green thing here is the is the cancerous tumour which needs to be a sector
0:16:48	so you have to identify
0:16:50	yeah the individual segments uh in the labour
0:16:53	the images themselves are actually displayed that us to do some i transparent goggles we the surgeon is uh is
0:16:58	varying
0:16:59	and that
0:17:00	the uh
0:17:01	oh
0:17:02	this is which you see here is really interactive uh panel which allows you to project the individual a sections
0:17:08	all dct in the proper orientation
0:17:11	uh
0:17:12	by all this place through the uh
0:17:14	sort of got
0:17:16	so an example of the uh
0:17:19	of the best three here you can identify and of you portions sort the vascular tree
0:17:23	and that as a result of that that you can get that
0:17:26	yeah segmentation all the ever in the segment because when you want to re sec portion of it we have
0:17:31	to recite the entire segment
0:17:33	you have to do the segmentation
0:17:35	as you will see uh
0:17:36	C immediately
0:17:38	uh uh after this
0:17:40	and uh a a a you can
0:17:43	you can use it then actually quite useful for the more difficult the a a sections
0:17:47	than uh do ones which are so really T
0:17:51	you uh
0:17:52	as get the lost the last
0:17:54	second of those in you light sleep segments
0:17:57	once you have that you can start to
0:17:59	uh
0:17:59	starting planning for the for the
0:18:01	and the are sections
0:18:03	you can get a a or was in the segments gets a i see that
0:18:06	to M or
0:18:07	i know that the oldest this particle segment has to go
0:18:10	so what sort of surgery a just take it out
0:18:13	and that
0:18:15	find out what happens if you do that do we have enough safety margin between the two
0:18:23	oh
0:18:24	such should uh
0:18:26	have to this uh a short introduction
0:18:28	it will consist of five different uh uh a papers
0:18:31	it a uh try to give you all some overview of the entire chain of typical operations steps
0:18:35	and the first uh a present they should be given bit by michael ones are beginning with the image reconstruction
0:18:41	interior tomography from by medical applications be the topic of the second talk
0:18:46	uh image and signal processing challenges in the translational uh likely imaging research
0:18:51	a given by all really if felt
0:18:53	then a talk above field you signals for respiratory gating long C image sequence
0:18:57	by uh joe right hard and last one be
0:18:59	how we can model at the origin a the a populations
0:19:02	given by
0:19:04	so this was a very brief introduction i think we are
0:19:07	the end of my a all the time
0:19:10	and that
0:19:11	if you have a one quick question i can answer or otherwise you problem be able to as a question
0:19:15	get more detailed uh
0:19:20	or right that's why whole
0:19:22	right
0:19:22	right

SIGNAL PROCESSING IN MEDICAL IMAGING AND IMAGE-GUIDED INTERVENTION

Medical Imaging

Přednášející: Milan Sonka, Autoři: Milan Sonka, The University of Iowa, United States