ROUNDTABLE WITH CHAIRS OF IMAGING-INTENSIVE DEPARTMENTS

Name: ROUNDTABLE WITH CHAIRS OF IMAGING-INTENSIVE DEPARTMENTS
Uploaded: 2025-10-27T17:25:59.3433333Z
Duration: 1 h 2 min 25 s

October 27, 2025

Information

ID: 13560
To Cite: DCA Citation Guide

Download Transcript

00:00I'm gonna go by alphabetical
00:02order so that,
00:03that's what I have in
00:04my list. So doctor Peter
00:06Glaser is the Robert Hunter
00:07professor of therapeutic radiology and
00:09a professor of genetics,
00:11and chair of therapeutic radiology.
00:13And I have to say
00:14he's been my mentor when
00:15I first came here. He's
00:15my my buddy, and he's
00:16been extremely helpful to navigate
00:18the Yale,
00:19system.
00:21Doctor Wolfram Gosling, who just
00:22joined only recently,
00:24and I know him a
00:25lot more here than when
00:26we were both for twenty
00:27five years,
00:28in a small clinic in
00:30Frits Street, MGB.
00:32He's the,
00:33enzyme professor of medicine and
00:34professor of cellular molecular physiology,
00:36and he's the chair of
00:37internal medicine.
00:38Pooja Hathri, who is the
00:40Albert Kent professor and chair
00:41of neurology,
00:42And, doctor John Crystal, who
00:44is the Robert McNeil,
00:47junior professor of translational research,
00:49professor of psychiatry, of neuroscience
00:52and psychology, and he's chair
00:53of psychiatry,
00:55co director of YCCI.
00:58Doctor Chris Whitlow, who has
01:00just landed after
01:01and he didn't break his
01:02foot coming here. It was
01:04on a set of stairs
01:05that he did that. And
01:06it's not because of us.
01:07He's been here at chair
01:08for only a day.
01:12He is the chair of
01:13the, Department of Radiology and
01:14Biomedical Imaging and assistant dean,
01:16for translational research.
01:18Doctor Eric Weiner, who's also
01:20I got to meet a
01:21lot more here than when
01:22we're both in Boston, and
01:24who is the Alfred Gilman
01:25professor of pharmacology,
01:26professor of medicine, medical oncology,
01:29deputy dean for cancer research,
01:30director of the ALCAN Center,
01:32and he's the president of
01:33the and physician in chief
01:34of the Smilow Cancer Hospital.
01:36And last but not least,
01:38Doctor. Yoshino O'Hoonomanchado
01:39couldn't make it, so
01:41she has put, doctor Hua
01:43Xu on call, and he
01:44had to show up. Thank
01:45you, Hua, last minute.
01:47He is the Robert McCloskey
01:49professor of biomedical informatics, data
01:50sciences,
01:51and vice chair for research
01:52and development. He's also the
01:54associate dean for biomedical informatics
01:56at Yale School of Medicine.
01:57So what we ask all
01:58of them is to start
01:59with
02:01five, six minute that they
02:02could tell us what works,
02:04what are the missed opportunities,
02:05what are the things we
02:06should be doing, especially in
02:07the research level, but also
02:09on the clinical side, And
02:10then we'll have all questions
02:11for them. And please don't
02:13be shy. I I know
02:14these may be your chairs,
02:14but this is the time
02:16to ask questions.
02:27It's
02:30alphabetical, I guess.
02:32That's fair. So hi, everybody.
02:34I thought I would just
02:36give a little bit of,
02:37overview of how we use
02:38imaging and radiation oncology and
02:40then maybe talk about some
02:42future directions
02:43because not everyone may be
02:44familiar with how things work
02:46in our department down in
02:47the basement over in Smilow
02:49and elsewhere.
02:50So,
02:51besides
02:52imaging for diagnosis and staging,
02:54of course, we,
02:56I think,
02:57uniquely integrate imaging in our
02:59treatment planning and our treatment
03:01delivery.
03:02So,
03:03we use
03:04CT, MR,
03:07PET,
03:08and
03:09even ultrasound sometimes
03:11as well as something called
03:12surface imaging,
03:15to,
03:16acquire
03:17images
03:18and to evaluate the extent
03:20of the malignancy
03:22as well as to identify
03:24healthy areas and health healthy,
03:26tissues and organs
03:28so then we can do
03:29our, treatment planning.
03:31And,
03:32and we do we start
03:34besides diagnostic tests, we have
03:36our own CT simulator,
03:38and in some departments, we
03:40have,
03:41people have MR based simulators.
03:44The CT simulators
03:45that we have not only
03:47take a standard static CT,
03:48but we can do time
03:49lapse
03:50imaging so we can follow
03:53motion,
03:55in in time so we
03:56can account for movement of
03:57tumors
03:58as we do our treatment
03:59planning.
04:00Now I've heard some people,
04:03in the field talk about
04:04what we do
04:05as a version of robotic
04:07surgery, but broken up in
04:08time.
04:09So instead of with the
04:11robotic surgery where you were
04:13looking at the,
04:15the the,
04:16monitor and you're moving the
04:17robot,
04:18to do the surgery in
04:20real time,
04:21We're using all those images
04:23to map out,
04:25the lesion,
04:26the areas to avoid in
04:28three dimensions in a very
04:30complicated,
04:31which I've also heard called
04:33fancy coloring book approach,
04:36to
04:37then allow
04:40our treatment planning systems to
04:42design,
04:43the treatment. And the treatment
04:45happens with our linear accelerators,
04:46which are dynamic.
04:49They move, they rotate,
04:51the aperture opens and closes.
04:53So the delivery of the
04:54radiation beam
04:56can really be exquisitely shaped
04:57so we can deliver dose
05:00very precisely
05:02in a very shaped
05:04way. Now in order to
05:05do that, we have to
05:06account for positioning and motion
05:08management, and that's where imaging
05:10comes into
05:11play again.
05:13So, for a lot of
05:14patients before we treat, the
05:15moment before we treat, we
05:16do another CT scan because
05:18we have CAT scan
05:20built into the LINACs, and
05:22some LINACs also have MRs
05:23built in now,
05:25so that we can image
05:27that patient and make sure
05:28that that positioning aligns with
05:30how we set up the
05:31treatment.
05:32We also have a number
05:33of devices for mobilization,
05:36including
05:37masks and cradles and even
05:39for intracranial things we do,
05:42head frames that are screwed
05:44into the skull.
05:46More recently, we've
05:48been able to obtain a
05:49a PET based LINAC, and
05:51Pam mentioned this earlier.
05:52And so we now can
05:54use,
05:55a PET signal,
05:56and the, linac will acquire
05:58that in real time and
05:59then shoot the treatment beam
06:01back so we can do
06:02motion management
06:04through the breathing cycle and
06:05other things that happen,
06:07in real time and account
06:09for that motion, but also
06:11it allows us to shape
06:12the beam and the treatment
06:13even more precisely.
06:16So
06:16going forward in the future,
06:18we're looking to incorporate into
06:19that procedure
06:21novel pet tracers that not
06:22only allow us to follow
06:24the anatomy
06:25and the motion, but tell
06:27us something about the biology.
06:28For example,
06:30David Carlson in our apartment
06:32and others have used a
06:33marker of hypoxia,
06:35which is fluorine labeled mesonidazole,
06:38which accumulates in hypoxic regions
06:40of tumors.
06:41And in theory, that would
06:42allow us not only to
06:43track the tumor, but to
06:44do what we call dose
06:46painting so we could deliver
06:47higher doses to hypoxic regions,
06:49which are more resistant,
06:51to tumors.
06:53The the other thing we
06:54would like to do in
06:55partnership is, build on the
06:57advances in theranostics
06:59that are going on,
07:01not only because it's another
07:03version of radiation therapy that's
07:05systemic and and we're very
07:06interested in that, but also
07:08to think about how we
07:09combine external beam radiation therapy
07:11and systemic radiation therapy.
07:13And one area that I
07:15think is,
07:16right for further study is
07:17how that dosimetry is figured
07:19out when you're combining both
07:21external beam and and theranostics.
07:24The the other thing that
07:25we're interested in, you know,
07:27our department has a lot
07:28of drug development research going
07:30on. Part of that is
07:31drug delivery.
07:33We've partnered with biomedical engineering
07:35as well as,
07:37other departments, MBMB,
07:39to develop novel antibodies or
07:40tumor targeted peptides,
07:42which can be the basis
07:43for novel,
07:45radioligands
07:46and and, you know, we're
07:48also using that for cytotoxic
07:50chemotherapy and things like that.
07:52But one of the things
07:52we've done with that that
07:54that also
07:55is a form of theranostics
07:56is,
07:58and we had some NIH
07:59funding for this. We use
08:00one of those tumor targeting
08:01peptides to deliver
08:03nanoparticles containing gadolinium.
08:06So that could be combined
08:07with MR imaging, of course,
08:10but gadolinium
08:11has the property that when
08:12you hit it with x
08:13rays of certain energy, you
08:15knock off electrons and it
08:17generates in a transition state
08:18something called Auger electrons,
08:20which then add more bang
08:22for the buck in the
08:22radiation,
08:24delivery. So there's a number
08:25of areas where I think
08:26we can
08:27leverage,
08:28a lot of what's going
08:29on in the imaging program.
08:34So I'll stop there. Yeah,
08:35Al.
08:55Yes. That would be very
08:56interesting. And I think that,
08:58you know, with something that
08:59we're interested in in developing
09:01that program, As you know,
09:02we're getting into cardiology not
09:04just with restenosis but in
09:06V TAC.
09:07And actually, that peptide that
09:09I talked about that delivers
09:10to acidic tumors also can
09:12deliver to ischemic regions, So
09:14whether that could be of
09:15use is something we can
09:16talk about.
09:18I'll I think we should
09:19talk about this later because
09:20we just finished the study
09:21in pigs,
09:22Felicitas Pichari, who who you
09:24know,
09:25where they we had done
09:26membrane potential mapping
09:28in during the treatment for
09:30ventricular
09:31tachycardia to confirm where you're
09:33treating exactly.
09:34So this may be something
09:36of use.
09:41I think it's me who
09:44is next, and I I
09:45feel a bit like an
09:46imposter AI. I've only landed
09:48here shortly, and I don't
09:49even do imaging.
09:54And many of you actually
09:55do it. And so I
09:56think from a,
09:58you know, Department of Medicine
10:00perspective, we're clearly
10:02major users of of of
10:04both diagnostic
10:06and, imaging and and, you
10:08know, both therapeutic
10:11intervention and and and application.
10:15You know, that is not
10:16to say that my colleagues
10:18in
10:19cardiology
10:20and gastroenterology
10:21do,
10:23live imaging as as they
10:24perform their procedures and and
10:26and,
10:27and diagnostics.
10:30So I think from from
10:31my perspective, what I'm excited,
10:33though, is on the on
10:35on what and what we
10:36need. And I we've heard
10:37some of this as we
10:38think about, you know, both
10:40chronic inflammatory disease and I
10:42won't talk about cancer, Eric,
10:43so I'll leave that also
10:45to you.
10:46You know,
10:47both chronic inflammatory diseases and,
10:51and and fibrosis, which affects
10:53so many,
10:54different organ systems and what
10:56we've heard where we can
10:58actually do functional imaging and
11:00and and diagnostics to really
11:01think about the molecular basis
11:03of these,
11:04processes and then,
11:07understand the,
11:09the the biological mechanisms of
11:11disease.
11:13You know, I I think
11:14microglial
11:15imaging is is fantastic.
11:17I think macrophage imaging in
11:19the inflamed liver or or
11:21as you highlighted in in,
11:23in in the lung is
11:25is, I think, of of
11:26of really both
11:29diagnostic importance for,
11:31for the different diseases, but
11:32also ultimately to see whether
11:34we can develop,
11:36novel therapeutic approaches. And so
11:38I'm excited to be here
11:40and to learn, but I
11:42and to see what all
11:43the imaging pros have to
11:44offer, and and then I
11:46can also refine my, my
11:48request of what I want.
11:56Okay. So, I can't use
11:58the excuse that I'm the
11:59newest chair anymore. I'm actually
12:01the third newest in this
12:02group, but I'm only six
12:04months in, so keep that
12:05in mind. But, you know,
12:07I actually took some time
12:08to just like look at
12:10imaging is so rich in
12:11neurology. It's such an important
12:13part of our research as
12:14a discipline.
12:16And,
12:17kind of went through, actually
12:18got Georges to, to make
12:20a list for me. And
12:20then also when an NIH
12:22reporter,
12:23we have over fifteen million
12:24dollars in NIH funded imaging
12:26research, and that doesn't include
12:28other, you know, foundations and
12:30sources of funds. So we're
12:31pretty busy and it's, it's
12:32a really exciting environment that
12:34I'm, I'm happy to be
12:35inheriting
12:37And it kind of hits
12:38the full range of things.
12:40And I'll just give you
12:41a few examples. You've heard
12:43some of them today, you
12:44know, for example, Carolyn Frederick's
12:46talk,
12:47and others, but, you know,
12:49we've got everything from like
12:51bread and butter localization
12:53of auditory
12:54awareness using fMRI
12:56and to hopefully plan neuromodulation
12:59interventions.
13:00We've got looking at
13:06mechanism for why we see
13:07cognitive delay,
13:09or impairment in our patients
13:10with HIV that are otherwise
13:12well treated. And similarly,
13:15and that's on PET. And,
13:16and Carolyn had mentioned to
13:18you her work and sex
13:20specific factors that impact TAU
13:22pathology.
13:23So, so that's PET.
13:25And then, and then there,
13:27there's, you know, a whole
13:28bunch of stuff going on
13:29with MRI and biomarkers.
13:33There, there's work with,
13:35specifically looking at distinct biomarkers
13:37for predicting stroke recurrence, like
13:40vessel wall inflammation.
13:43There's a project that I've
13:44brought over to Yale, which
13:46is looking at, it's a
13:47biomarker validation study, looking at
13:50corticospinal
13:51tract lesion load,
13:53along basically plain old MRI
13:55that you get clinically, a
13:56poor man's DTI, and hopefully
13:58being able to implement that,
14:00you know, nationwide as part
14:01of our clinical tools. So
14:03we'll see about that.
14:05And then something that, you
14:06know, didn't hit the radar
14:08today, but probably good for
14:09people to know about is,
14:11members of my department have
14:12helped build a portable MRI.
14:15It's a low field MRI
14:16that really gives you some
14:17opportunity, especially for your patients
14:19with intensive care units, or
14:21maybe even more mobile locations
14:22to get access as you're
14:24thinking about access to care.
14:27So that's kind of an
14:28overview, you know, in really
14:29big picture terms about what
14:31my department is up to.
14:33I'll say I'm really excited
14:34that we're going to be
14:35getting at the ADAMS neurosciences
14:37tower and it's going to
14:38have the PET MRI in
14:39it.
14:40So, so that's a big,
14:41I think, exciting win for
14:43us.
14:44And, and then I'll and
14:45then finally, I surveyed a
14:46few people this afternoon. I
14:48said, I'm supposed to talk
14:48about what we, we hope
14:50we could do better.
14:52And,
14:53I got a bunch of
14:54random things, but I'll say
14:55there was one theme for
14:57us, which was people who
14:59are doing clinical research that
15:00rely on using, you know,
15:02things like clinical MRI, where
15:04we're just running into so
15:05much, so many problems with
15:07what is a nationwide shortage
15:08in radiologists and radiology techs,
15:11and feeling like there are
15:12great ideas, but sort of
15:13hitting a wall in terms
15:14of access for our clinical
15:16research. So I'm sure that's
15:17not a surprise to many
15:18people here, but that certainly
15:20was top of mind for
15:21a lot of my colleagues.
15:24So, Puja, could I ask
15:26a question related to this?
15:27So,
15:28and maybe for all of
15:29you, as you saw, there's
15:31quite a bit of equipment
15:32that is mostly under NIH
15:34funding,
15:35but that is being planned
15:36to be installed.
15:38And one of the things
15:39we have been toying with
15:41the idea, and a lot
15:42of my constituents are gonna
15:43kill me after I say
15:44this, but I'll say it,
15:45is why not think of
15:47a one day a week
15:49giving access to clinical use
15:50for some of the research,
15:52especially something like a seven
15:53Tesla or a PedamR?
15:56Is this something all of
15:57you would consider, or is
15:58this would be a nightmare
16:00if this was built
16:01in the correct setting that,
16:03you know, with access to
16:04patients, assuming all of that
16:06is done? Is that something
16:07you'd consider, or is it
16:08gonna be such a nightmare
16:09that no thanks? Are you
16:10talking about sharing clinical imaging,
16:13with or research imaging the
16:14other way around? Yeah.
16:16Only like a day a
16:16week. Yeah. Yeah. Because I
16:18need to leave this room
16:19out. So but say a
16:20day a week.
16:22Do others wanna comment? I
16:23I'm just gonna say I'm
16:24I'm a big fan of
16:24sharing and and and using
16:26things effectively. So if we
16:28cannot take away from others
16:29and and increase the capacity,
16:31I don't see why we
16:32wouldn't do that. Yeah. I
16:33mean, I would say from
16:35from the point of view
16:35of using advanced imaging as
16:37a radiologist
16:39that,
16:40it's kind of one of
16:41these things where we we
16:42don't know what we don't
16:42know as as a radiologist,
16:44and we don't know what
16:45the advanced tool you know,
16:47we we have kind of
16:48clinical grade equipment,
16:50that's not state of the
16:51art, and we know what
16:52that produces because we read
16:53those studies every day. But
16:55we don't really know but,
16:57you know, to put a
16:58high performance tool in our
16:59hand to just sort of
17:01you know? And I don't
17:01have a good way to
17:02phrase it, but just sort
17:03of play around with it.
17:04And I know that doesn't
17:05sound very sophisticated, but to
17:06play around with the tool
17:08to see
17:09what's possible, you know, what
17:10what kind of,
17:12what kind of nuanced
17:14image it can produce that
17:15helps us make a better
17:16diagnosis, I think, would be
17:17helpful. And it's just it's
17:19just really hard to predict
17:20what that would be, and
17:21therefore, there's not a lot
17:22of support for it because,
17:23you know, it's not really
17:24hypothesis driven. It's just more
17:25of, like, literally, it's just,
17:27like, kind of play. You
17:28play around with it and
17:28see what it's good for.
17:29But in order to do
17:30that, you have to have
17:31access. So, yeah, it's it's
17:33a good point. I think
17:34there would be some utility,
17:36and we might learn how
17:37to leverage maybe seven t
17:39or even advanced three t
17:41or explore type,
17:43resolution
17:44better to the advantage of
17:46our patients.
17:47And don't you see also
17:48a value in not only
17:50giving access, but also
17:53training physician scientists, attracting residents
17:56for our own,
17:58you know, attending to having
17:59something
18:00exciting
18:02other than the bread and
18:02butter. Because I heard you
18:03about the bread and butter,
18:04but this is, like, something
18:05that would be really exciting.
18:06If it's if it's being
18:08used clinically in a short
18:09period of time, then it
18:10takes that barrier of, like,
18:11yeah, but this is research
18:12we can't use in our
18:13patients. Do you see value
18:14in that?
18:18Sure.
18:21Yes.
18:24I said yes.
18:27Who's this one say more?
18:28I mean, do you have
18:29any is the new
18:31I'm trying to it's the
18:32same same. It's stick this.
18:34Yeah. It's it's it's just
18:36the same. Like, you you
18:37you need to have the
18:38tool in your hand
18:39to be able to and
18:40then to apply it to
18:42see what it's good for.
18:43And it's kinda one of
18:44those things where you don't
18:45you don't know until you
18:46use it. And then perhaps
18:47there's some relationship that you
18:49wouldn't have seen with a
18:50conventional clinical scanner that you
18:51can see with a more
18:52advanced scanner. And, yes, it
18:53would it would make, it
18:54would excite the residents and
18:55excite the trainees,
18:57and there'd be lots of
18:58ways to leverage that. And
18:59and there's important, you know,
19:01maybe not hypothesis basic science
19:03research that comes from that,
19:04but more observational type research
19:06that can also be impactful
19:07to patients,
19:08which which the residents could
19:09drive and trainees could drive,
19:11and and a lot of
19:12our clinical faculty clinically oriented
19:14faculty
19:23Well, I just just follow-up
19:25on that. I, you know,
19:26I mentioned we use MR
19:27for treatment planning. I think
19:29some of these advanced tools
19:30potentially we could incorporate into
19:32the treatment planning because it's
19:33not used diagnostically,
19:35so we don't have that,
19:37you know, barrier to getting
19:39approval or, you know, reimbursement.
19:41But we're always looking for
19:42better tools to guide the
19:43treatment, and that could, for
19:44example, be baked into a
19:45clinical trial or something like
19:47that.
19:52Do you want me to
19:53go?
19:54All right.
19:55So,
19:57you know, before neuroimaging,
20:00psychiatry had,
20:02three basic pieces of equipment,
20:05comfy chair, a pad of
20:07paper,
20:08and a writing instrument.
20:11And,
20:12I say that,
20:14because,
20:16it's hard to capture how
20:18profound the change in the
20:20field of psychiatry and how
20:21profoundly psychiatry Yale has changed
20:24since the advent of neuroimaging.
20:27And it's been extraordinarily
20:29impactful.
20:30It has,
20:31changed the discussion,
20:34discussions that people have at
20:36all levels of training and
20:37in clinical practice.
20:40And we've been extraordinarily
20:41fortunate at Yale to have
20:43just amazing,
20:45neuroimaging
20:46resources to draw on that
20:47has,
20:48enabled really impactful research,
20:51to emerge, and also
20:53exciting new kinds of collaborations,
20:56both within Yale and,
21:00collaborations with, pharmaceutical industry, biotechnology,
21:03and other kinds of things.
21:06I thought I would make
21:08three points and then I
21:10can,
21:11we, I can address, concerns
21:13and things like that.
21:15So in psychiatry, we've used
21:18it across
21:19a wide range of purposes
21:21To, understand the biology of
21:23basic cognitive and behavioral mechanisms,
21:26characterizing
21:27individual differences in biology across
21:30development, degeneration,
21:33understanding illness related pathophysiology,
21:35developing new treatments,
21:37and developing biomarkers to guide
21:39personalized treatment.
21:41We've used in psychiatry
21:43pretty much every neuroimaging modality
21:46that we have here at
21:47Yale.
21:48And I would say that,
21:51one of the things that
21:52has characteristically
21:53distinguished Yale neuroimaging, which interestingly
21:56hasn't
21:57come up in
21:58such a clear way today,
22:00is Yale's really good not
22:02only in making,
22:04static measurements,
22:06the expression of a protein,
22:08the,
22:11volume of a region,
22:13but in both in the,
22:15spectroscopy
22:16work and in the PET
22:18work has been particularly
22:20good at characterizing
22:21dynamic processes. The release of
22:23a neurotransmitter,
22:25how that relates to other
22:27cellular functions. And that,
22:29that innovation has been extremely
22:31important for the kind of
22:32impact for work that has
22:33emerged,
22:34at Yale and in psychiatric
22:36research.
22:39The other thing I would
22:40say, and I think I
22:42thought that
22:43in psychiatry, but in the
22:44general session today,
22:47about the role of neuroimaging
22:48or imaging as a bridging
22:50technology
22:51between,
22:52it's one of the only
22:53ways you can get the
22:54same molecular finding in brain
22:56in an animal and a
22:57human.
22:58That's fundamentally
23:00important across all areas
23:02of,
23:03biology and true,
23:06in psychiatry as well.
23:08You can make the same
23:09measurement in post mortem tissue
23:11and in vivo tissue. That's
23:13extremely
23:14important because you can only
23:16work out what the disease
23:17is in the organ that
23:19the disease has. And usually,
23:21in people, we don't get
23:22to study the dis the
23:24organ that,
23:25of the,
23:27the brain,
23:28access is obviously very challenging.
23:31It's a great way to
23:32link computational model to to
23:35to in vivo,
23:36data and validate and and
23:39update those models.
23:42And,
23:45and I think the major
23:46challenge that we have in
23:47psychiatry
23:49is that
23:50neuroimaging, although we've been doing
23:52neuroimaging research
23:54for
23:57thirty
23:58years,
24:00it's still a research tool.
24:02We're not,
24:03characterizing
24:04D2 receptor occupancy by an
24:06antipsychotic in order to adjust
24:08antipsychotic
24:09dose. We just
24:10look for the side effects
24:12and and adjust the dose.
24:15And,
24:16so, you know, thinking about
24:18ways that we can
24:20turn molecular,
24:23molecular,
24:26mechanisms into
24:29biomarkers that guide treatment is
24:30still an emerging
24:32challenge and something that a
24:33number of people in the
24:34department are working on. And
24:36the model
24:37is probably Alzheimer's disease, where
24:39you have have a a
24:41a barren protein that you
24:43characterize and then you target
24:44that protein,
24:47or an aberrant
24:49circuit activity,
24:51that you might target with,
24:52say, neurostimulation
24:53treatment or or,
24:56other kind of intervention.
24:59So
25:01what do we want?
25:03We want we want more,
25:04and we want it all
25:05the time, obviously, George, and
25:07for for less money, of
25:09course.
25:11You know, we have
25:12among the most,
25:14rich array of ligands to
25:16probe different targets. Of course,
25:18there are critical targets that
25:20are relevant that we that
25:22we haven't yet gotten to.
25:24And and,
25:25really, the targets I think
25:26SV two a is a
25:27great example of a of
25:28a
25:29protein target
25:31that
25:32that was ahead of the
25:33field when we identified it.
25:35Congratulations to
25:37Rich.
25:38And,
25:39and,
25:40and brought the field forward
25:43to integrate.
25:44And so we want to
25:45keep pushing the field
25:47to find those,
25:50those,
25:53ligands that help us to
25:55change the science, really. And
25:58this is an exciting time
25:59to do it because the
26:01ability to move back and
26:02forth across levels of analysis
26:04is so great. The other
26:06thing I think
26:07we may get with a
26:09PET MR
26:10is
26:11dynamic measures at the same
26:13time as we're making molecular
26:14assessments. And I think that's
26:16a really exciting
26:18opportunity as well. You know,
26:20we've messed around, those of
26:21you who've done this, know
26:23we've messed around with interleaved
26:25EEG
26:25and and and MR, and
26:27that's been
26:30interesting.
26:32But this this, I think,
26:34is is more is more
26:35promising and and,
26:38will guide the science. So,
26:40you know, mostly,
26:42I'm here to say thank
26:43you
26:44to Yale neuroimaging
26:46because
26:47it's really driven,
26:49kind of culture of innovation
26:51in psychiatry that has been
26:52extremely valuable for
26:55making us a a a
26:56good department. So so thank
26:58you.
27:05Go ahead.
27:06Oh, either way. Either way.
27:09Thanks for thanks for coming.
27:12In terms of what imaging
27:13does well and then what
27:14are the challenges,
27:16in particular with MRI, you
27:17know, we've heard a lot
27:18of great
27:19about a lot of great
27:20MRI research. And,
27:23MRI scanners are interesting because
27:24when they're being built,
27:27there are these measurement tools
27:28that are,
27:30on the factory floor,
27:32tuned and shimmed
27:34to optimize,
27:35tissue contrast
27:37specifically for human eyes so
27:39that we, as radiologists,
27:41can see pathology. And it
27:43and it does a great
27:44job at that. So if,
27:45let's say, a patient comes
27:46into the emergency room, they
27:48eventually get in maybe for
27:49epile seizure or something, and
27:50they get an MRI. And,
27:52we do a brain scan,
27:53and then I see this
27:54giant
27:55heterogeneous
27:57enhancing lesion with cystic and
28:00solid components and restricted diffusion
28:02is causing mass effect and
28:03pushing the brain all around.
28:04And,
28:06and and maybe they're in
28:07their fifties. And, and then
28:09I'm I'm thinking, okay. Well,
28:10they've got some sort of
28:11aggressive tumors, probably glioblastoma, and
28:13then I send them on
28:14their way to get, you
28:15know,
28:16part of that cut out
28:17and then doctor Glaser to
28:19do some radiation and etcetera
28:21to treat it.
28:22But, you know, there's this
28:24great paper by,
28:26Hadi Risak, who is chair
28:27of radiology at Memorial Sloan
28:29Kettering, and I think I
28:29can't remember the exact title,
28:31but it it's something like,
28:33medical images are more than
28:35just pictures. They're data. So
28:37the the images that I'm
28:38looking at,
28:40qualitatively to say there's this
28:42big thing here,
28:43there's data in there at
28:44at the at the voxel
28:45wise level, and a voxel
28:46is just a three-dimensional pixel.
28:48So if you, you know,
28:49if you look really closely
28:50at your computer monitor, you'll
28:51see that it's made of
28:53of a bunch of dots,
28:54like pointillist
28:55paintings,
28:57and apps impressionist paintings. But
28:59they're three-dimensional,
29:00voxels. And within each voxel,
29:02there's there's data there. There's
29:03quantitative data,
29:05and that can be and
29:06so these these pictures can
29:08be processed in a way
29:09to extract this quantitative data,
29:10and that's what radiomics
29:12is.
29:13And there's a bunch of
29:14papers about how radiomics can
29:16do all kinds of things
29:17for brain tumors and predict
29:18predict
29:19outcomes and all of this
29:20all of these great things,
29:21but, you know, we're not
29:22using it in the clinic.
29:24And you might say, well,
29:25why is that? Like, why
29:26aren't why aren't you know,
29:27there's all this data to
29:28suggest that radiomics are great
29:30and they can make all
29:31these predictions.
29:32Why are why why do
29:33my reports still just describe,
29:35you know, this this,
29:37in a qualitative way, this
29:38big tumor? Why isn't it
29:39discussing the radiomics features that
29:41we could extract? And that's
29:42because the thing that MRI
29:44does bad is that it's
29:45it's it's,
29:47when they when they engineer
29:48these scanners,
29:49they're engineered in a way
29:50that, again, they're tuned for
29:52our eyes, but it it
29:54makes a measurement tool that,
29:56for quantitative
29:57purposes,
29:58there's massive
30:00measurement error between scanners. So
30:02for example, let's take something
30:03like, you know, you saw
30:04some,
30:05fractional anisotropy
30:07diffusion images of the brain,
30:09these beautiful
30:10colorful tracks of of white
30:11matter in the brain, and
30:13you can measure you can
30:14make measurements of of those
30:15tracks, things like fractional anisotropy.
30:17And so let's say you
30:18took,
30:20doctor Brown and you put
30:21her in an MRI scanner
30:23here,
30:24and and they can be
30:24they can be two scanners.
30:25They can be same same
30:27vendor,
30:28built side by side on
30:30the factory floor, installed at
30:32the same institution at the
30:33same time,
30:34same,
30:35hardware, same software platform. Everything
30:38is, like, nearly identical. You
30:40can scan her brain. You
30:41can make a FA measurement
30:43and in one scanner and
30:44then scan her immediately in
30:45the next scanner.
30:46The measurement will be thirty
30:47percent different thirty percent different.
30:49So imagine if you were
30:50trying to, you know, take
30:52someone's weight
30:53and you're trying to track
30:55someone's weight, and it's thirty
30:56percent different between two scales.
30:59So MRI scanners are great
31:00at making measurements,
31:02but,
31:03between
31:04instruments, there's there's a problem.
31:06And and that's why that's
31:08why, you know, you can't
31:09reliably extract all of these
31:11wonderful quantitative metrics
31:13because, you know, if someone
31:15is scanned here and then
31:16they're scanned for follow-up at
31:17another at another institution, these
31:19aren't things that you can
31:20track between scanners over time.
31:22So then, you know, that's
31:23where AI comes into play.
31:25You know, you can't talk
31:26about imaging without AI.
31:28And,
31:29AI is interesting because it's
31:31not limited, number one, by
31:32our senses. So it can
31:34it it can recapitulate
31:35a lot of things that
31:37HI can do. So AI
31:38can do things HI can
31:39do, so human intelligence. So
31:41AI can say there's there's
31:43a tumor, and I can
31:45say there's a tumor. Well,
31:45that's okay. So AI can
31:47do the thing that I
31:47can do. But AI, I
31:49think where it's really interesting
31:50is where it can do
31:50things that we can't do.
31:52And,
31:54and one of those things
31:55might be to
31:57overcome that hurdle of the
31:58engineering of MRI scanners to
32:00maybe
32:01harmonize,
32:02MRI scanners
32:04across each individual measurement device
32:07so that we can unlock
32:08these quantitative
32:10measurements and then use them
32:11to great effect. And then
32:12also do things like maybe
32:14even leverage functional measurements like,
32:15you know, you've heard you
32:17saw,
32:18you know, that there are
32:19changes in default mode network
32:20in Alzheimer's disease patients. Well,
32:22why are we not using
32:22that clinically? Again, because there's
32:24heterogeneity
32:25across
32:26MRI scanners, and what we
32:27need to do is harmonize
32:28them. And this has been
32:29done with CT. So if
32:30you take a CT scanner,
32:32CT scanners are,
32:34calibrated to water. So water
32:36is zero density or, you
32:38know, zero units,
32:39and then fat is negative
32:41because it's less dense than
32:43water. Bone and muscle are
32:45positive.
32:46So, you know, you can
32:46you can scan someone at
32:48a c on a CT
32:49scanner here at Yale. You
32:50can scan them on the
32:51West Coast on a scanner
32:52that's five or ten years
32:53old, and you'll get a
32:54measurement that's within a reasonable
32:56range of of,
32:58of reliability and and,
33:00between the scanners. But, again,
33:02we're not there with MRI.
33:04That's a huge problem. And
33:05once we can do that,
33:06then that's going to unlock
33:08all kinds of ability to
33:09do quantitative things
33:11and, and then leverage all
33:13of that quantitative data, these
33:14huge clouds of quantitative data,
33:16and then apply
33:18things like AI on top
33:19of that to do things
33:20that humans can't do, like
33:21predictive analytics. And we already
33:23know that that the imaging
33:24data, let's say, in a
33:25brain tumor, you can the
33:27the imaging phenotype of a
33:28brain tumor can predict its
33:29genotype. So we know that
33:30there's data there in these
33:32in these images that are
33:33incredibly powerful and predictive if
33:35we can unlock it in
33:36a way that is,
33:38reliable
33:39and valid
33:41across scanners, across platforms.
33:43But, so it's a huge
33:44challenge. It's one that my
33:45lab is working on and
33:46many people are working on.
33:48So I guess we'll see
33:49what the future brings. But
33:52So before we move, I
33:53think it would be a
33:54miss if, I don't know
33:54if Gigi Galliano is in
33:55the room, but she is
33:56working on, Gigi, are we
33:58here?
33:59Do you wanna say something?
34:00Because you're working on this.
34:06Yeah.
34:07So it would be interesting
34:08to talk. We did develop
34:10an interesting,
34:12approach to standardized MRIs,
34:15based on
34:16typical clinical acquisition. So it
34:18would be
34:19nice to share.
34:26Alright.
34:27So,
34:28I will briefly describe the
34:30activities at the Department for
34:32Biomedical Information and Data Science
34:34for BIDS.
34:35So at the BIDS, basically,
34:37we're working on all aspects
34:39of, informatics,
34:40data science, and AI
34:42ranging from data curation, data
34:45standardization,
34:46all the way to the
34:48model building,
34:49model deployment in the clinical
34:51settings.
34:52And we deal with all
34:53kind of data, clinical, genomic,
34:55as well as imaging data.
34:57Recently, a number of our
34:59faculties actually are working on
35:02building large energy model,
35:03as well as those multimodal,
35:05large energy model,
35:07involving imaging data.
35:09For example,
35:10Doctor. Qingyu Chen, assistant professor
35:13at BITS,
35:14He's,
35:15doing a lot of work
35:16on
35:17the multimodal,
35:19LLMs for the ophthalmologist
35:21data. So he collects, like,
35:23hundreds of thousands of ophthalmology
35:25images
35:26together with the text description,
35:29align them to a text
35:30vision large language
35:32model, and allow it to
35:34actually do,
35:35eye disease diagnosis as well
35:37as providing
35:38reasoning process.
35:40And after developing model, he
35:42also actually
35:43applied
35:44to, clinical studies, recruited, like,
35:47twenty four clinician
35:48across,
35:49twin
35:50twelve different medical centers.
35:52And through evaluation
35:54show when clinicians
35:56assist with AI greatly improve
35:58their
35:59diagnosis accuracy on eye disease,
36:02diagnosis tasks.
36:03And,
36:04besides,
36:07research, I think, the training
36:10next generation of medical AI
36:11expert is also a priority
36:13of, BITS.
36:14I know George, the imaging
36:17institution and,
36:18us, BITS are talking about
36:20potential,
36:21new joint training program in
36:23the medical AI and the
36:24imaging.
36:25And, the BITS team also
36:27provide,
36:28quite a lot of data
36:29science related services.
36:31I think many of you
36:32know the doctor,
36:34Daniela Micker
36:35leads the JEDA team, which
36:37actually provide,
36:38data retrieval service for all
36:40kind of data, including imaging
36:42data.
36:43In the last year, we
36:44also,
36:45worked extensively
36:46with the YCRC,
36:48with, HSIT,
36:49as well as the security
36:51office,
36:52to build a more,
36:54secure computing environment. So now
36:57we have two NIST eight
36:58hundred one hundred and seventy
36:59one environment ready. One is
37:02the Hopper, which is a
37:03GPU high end GPU cluster.
37:06The other one is a
37:07Spinrite Plus, which is a
37:08AWS based environment.
37:10Both of them meet the
37:11this eight hundred one seventy
37:13one environment.
37:15And,
37:17yeah, we we start to
37:18use it. So I just
37:19want to quote for the
37:21hopper. It's free to use
37:23until end of December.
37:26So go ahead and register
37:28and and become a user.
37:29After starting January, we're talking
37:31about fee structure for using
37:33the,
37:33it it has
37:35sixty,
37:36h one hundred GPU as
37:38well as a thirty two
37:39h two hundred GPU cards.
37:42And,
37:43regarding unmet needs, I I
37:44just quickly I think for
37:46us doing the AI modeling
37:48data
37:49and the computational,
37:50resource still two big challenges
37:52for us.
37:53I know we have a
37:54lot of data in the,
37:55sort of practice and the
37:56research,
37:58but, how to,
38:00obtain large,
38:01high quality,
38:03well annotated data still still
38:04a challenge for
38:06us. And, computational resource with
38:09the AI initiative at Yale,
38:10I think we are in
38:11a quite good position in
38:13terms of GPUs. There's a
38:15couple of high end GPU
38:16class available now.
38:19But I think the storage
38:20might be one,
38:22challenge because I involve a
38:23few large scale,
38:25imaging based study.
38:28They are using they are
38:29having data like hundreds of
38:31terabytes.
38:32We have two terabytes on
38:34the hopper. If we
38:36have three or or ten
38:37project like that, we we
38:39already full. So how to
38:40efficiently measure this kind of
38:42large scale
38:43imaging data, I think it's
38:44a it's a it's a
38:46thing we we should think
38:47about next.
38:48That's fine.
38:50Yeah.
38:55My turn?
38:56Okay.
38:57It's, you know, when you
38:58have a last name beginning
38:59with W, as Chris knows
39:01as well, you're often at
39:02the end.
39:05So,
39:08I think that,
39:11the first thing I would
39:12want to say is that
39:13I think that as a
39:14field,
39:15we are guilty of not
39:17taking sufficient advantage of imaging
39:20from a research perspective.
39:22And
39:23I think that that is
39:25it it's it's really quite
39:26profound.
39:27And if you go to
39:29our
39:30annual cancer meetings,
39:32there are hardly any presentations
39:35that really focus on imaging
39:38as a as a tool
39:40to help us better understand
39:42cancer.
39:44So in my mind,
39:46you know, we, of course,
39:47use imaging all the time.
39:50But imaging has a role
39:53theoretically related to well, not
39:55theoretically, in practice, related to
39:56early detection,
39:58to diagnosis
39:59of symptomatic disease,
40:02to monitoring
40:04response to therapy,
40:07delivering therapy,
40:09and then in follow-up.
40:11And it's interesting
40:13that over the past few
40:14years with the development
40:17of circulating
40:18tumor DNA that that's all
40:20the rage.
40:21But in truth, it's something
40:23that should have been developed
40:25hand in hand with imaging.
40:27And where imaging
40:29interdigitates
40:30in there
40:31very, very nicely.
40:35So let me just start
40:36with what I think is
40:38one of the most straightforward
40:40issues and then get a
40:41little bit more complex.
40:43So unlike Puja, I didn't
40:45think to ask any of
40:46my colleagues. I did, however,
40:48consult with,
40:50Google AI today.
40:53And
40:54I asked,
40:55probably a question that was
40:56just too complicated, which is
40:58that for the typical woman
40:59with metastatic breast cancer, I'm
41:01a breast cancer doctor, so
41:02that's always my go to,
41:05how many,
41:07CT scans does that woman
41:08have during the course of
41:09her illness? And, of course,
41:11it said to me that
41:12that was too complicated, and
41:13it had individualized and this
41:14and that. But, you know,
41:16in truth, the number is
41:17probably something like twenty during
41:19someone's life with metastatic breast
41:22cancer.
41:23We don't know a darn
41:24thing about
41:26the frequency we should do
41:27those scans,
41:29whether they should be more
41:30often, whether they should be
41:32less often, how we should
41:33interpret them, how they affect
41:35the quality of care. We
41:37know that they're certainly costly.
41:39We don't know to what
41:40extent they may,
41:42though, save money in some
41:44places and cost us a
41:45lot of, dollars at others.
41:48And very simple questions across
41:51all malignancies
41:52about how we should do
41:53imaging
41:54clinically and in research studies
41:57are worth asking, and we
42:00very, very rarely do that.
42:02So now let's get a
42:03little bit more complicated. So
42:05Pam talked earlier about theranostics.
42:08I was
42:09a big naysayer for a
42:11long time.
42:12You you did not you
42:13did not call me out.
42:15However, theranostics is very clearly
42:18here to stay, a useful
42:21modality.
42:22And the reason, of course,
42:23that many of us were
42:24naysayers is that the early
42:26attempts to do theranostics
42:28were highly, highly toxic. And
42:30it's only by through the
42:32development of,
42:34treatment approaches that are both
42:36effective and less toxic that
42:38that it has convinced many
42:40of us that this is,
42:41you know, really here to
42:43stay.
42:44But in terms of theranostics,
42:45we need help understanding
42:48how to dose these agents,
42:50how both in terms of
42:51the the the quantity
42:53per dose, how frequently they
42:55should be given, how long
42:57they should be given, all
42:58of these sorts of things.
43:00And that's what we really
43:01have to turn to you
43:02for.
43:03The other rage in oncology,
43:06antibody drug conjugates.
43:08God knows how many presentations
43:10I've heard about new antibody
43:12drug conjugates,
43:13virtually none of which have
43:14included an imaging component, and
43:16you would think that this
43:18is something that we could
43:20be able to image. Yes.
43:22Exactly.
43:23But,
43:24do we do we see
43:25that?
43:26No. And so we desperately,
43:29need it.
43:32And not just related to
43:34diagnostics and antibody drug conjugates,
43:36but with all of our
43:37new therapeutics.
43:38And, you know, in
43:40breast cancer alone, there have
43:41been fourteen
43:43FDA drugs or FDA approvals
43:46in the last decade.
43:48And, you know, you you
43:50multiply that times all of
43:51the diseases. We're talking about
43:53hundreds of drugs approved in
43:54the oncology space. And we
43:56need to better understand how
43:57to use all of those
43:59drugs. And again, it's both
44:01in terms of
44:02dose
44:03and in terms of particularly
44:05duration of therapy.
44:07And the default,
44:09in our field
44:11is to use
44:12doses that are too high
44:14to begin with and and
44:16and treatments that go on
44:18too long,
44:20resulting
44:21in both cost and,
44:24unnecessary
44:25toxicity
44:26and just, you know, general,
44:28waste.
44:30And then, finally, my own,
44:33you know, sort of pet
44:34peeve,
44:36not that I'm alone here,
44:38which is about intratumoral heterogeneity
44:40that I think we still
44:42only,
44:43you know, to have scratched
44:45the surface in terms of
44:46the extent to which that's
44:47a huge resistance mechanism in
44:50oncology.
44:51But, you know, this is
44:52another area where imaging could
44:54help us tremendously.
44:56So I I think that,
44:58there's a huge need for
45:00collaboration
45:01between,
45:03both
45:04basic cancer researchers and people
45:06who do imaging and very
45:08much, people who do clinical
45:10and translational work and in
45:12imaging.
45:12And I think one of
45:13the great things about this
45:17workshop today
45:18is the ability to get
45:19people together to talk.
45:22So
45:24Since I have you sitting
45:25here, can I ask you
45:26a question?
45:28So when for questions anyway.
45:31When you
45:33when you observe, right, how
45:34we have not been able
45:35to take advantage
45:37of,
45:39of some of these
45:40really fast and and excitingly
45:42moving forward
45:44imaging technologies in in the
45:46daily work that you do,
45:47that I do.
45:49Do do you think we're
45:50just in in clinical medicine
45:52too conservative?
45:55Because,
45:56you know, I'm I've been
45:57looking at liver
45:58scans for one reason or
46:00the other to to to
46:01detect early liver cancer for
46:03two and a half decades.
46:04And the protocol I used
46:05then is the same that
46:06I use now. And the
46:08the protocol that radiologists
46:10used then is the same
46:11that they use now. And
46:13so are we just then
46:15too content with what we
46:17have, or do our guidelines
46:18sort of lock us in
46:20into what we should do?
46:22And,
46:23sort of it's good enough
46:25for today. And so they
46:26therefore, it'll be good enough
46:27tomorrow. And what
46:29what is the push that
46:31we need to have,
46:32right, that,
46:34to move us forward. I
46:36was, rounding with one of
46:37our teams last week, and
46:39and you know what the
46:40biggest eureka moment was for
46:42all of you? It was,
46:44I didn't know we can
46:45do a portable lateral
46:47x-ray.
46:49And I no. I'm not
46:51kidding.
46:52Right? And
46:53I I'm like, they cannot
46:54be the innovation that we
46:55are talking about.
46:58So,
46:59yes, yes, and yes. And,
47:01you know, I have felt
47:02for a long time
47:04that what sometimes feels like
47:06very incremental progress that a
47:08lot of people are content
47:09with
47:10is because we're just too
47:12conservative. Now I think some
47:14of that comes from a
47:15good place, and it comes
47:16from the place of wanting
47:18always to do our best
47:19and
47:20by by people
47:22who are sitting right in
47:23front of us and not
47:24put them at any risk
47:26whatsoever.
47:28And, you know, we clearly
47:29need to do that. But
47:30at the same time, we
47:32do need to be far
47:33bolder. I think the other
47:35big challenge is we just
47:37tend to sort of stay
47:38in our silos and do
47:39the thing we do,
47:41and we gotta get out
47:42of it.
47:43Because, you know, without that,
47:45we're just not gonna make
47:47I mean, we'll make progress,
47:48but it's just too slow,
47:50too costly,
47:51you know, and and we
47:52can do much better.
47:54You know, I've got I've
47:55got a a kind of
47:56a question and a comment.
47:57Do you think do you
47:58think one of the I
47:59I noticed that in radiology,
48:01we also kind of stay
48:02in our silo. We learn
48:03to do something well, and
48:04then we just keep doing
48:05it and we repeat it
48:06because we know how to
48:07do it well, and it's
48:08and it seems to be
48:09helpful.
48:10Do you think that one
48:11thing that
48:13keeps us in our lane
48:14is just that physicians
48:16are so busy these days?
48:17Part part of part of
48:18why we keep doing the
48:19same thing is because, you
48:21know, there's enormous amount of
48:22work to do, and we
48:23feel lucky and and grateful
48:25every day just to get
48:26the work done
48:27at a reasonable time so
48:28we can go home. And
48:29there's just very little time
48:30to think about,
48:31you know, innovating.
48:33And, you know, and and
48:34and I worry a little
48:36bit about the physician scientist
48:39phenotype that
48:41that clinical work is so
48:43busy, and there's so much
48:44work to be done that
48:45there's not as much time
48:46to
48:47just be in your head
48:49and imagine what could be
48:51and ruminate on things in
48:52a way that leads to
48:53innovation. And I don't know
48:55what your thoughts are there.
48:56Well, I agree, but I
48:57think that's why we have
48:58Yale's.
49:00I mean, like, that's we're
49:01we're supposed to be doing
49:02that here. And I think
49:04that,
49:05you know, we have a
49:06lot of people who do
49:08pure clinical medicine in the
49:09system, but I think those
49:11people are all too willing
49:13to follow the lead of
49:15people who wanna do things
49:16a little bit more creatively.
49:17Yeah. And I think we
49:18just have to keep pushing
49:19ourselves. Yeah.
49:23Yeah. We can get your
49:24microphone up.
49:35So I wanna push back
49:36a little here on the
49:37lack of innovation.
49:40So you may have been
49:40looking at fatty liver disease
49:42for the last twenty five
49:43years and the fundamental physics
49:45of fat and water hasn't
49:47changed. So the basic pulse
49:49sequences haven't changed.
49:51But those images are probably,
49:52you know, two to three
49:54times the resolution
49:55that they used to be.
49:57They're probably collected in a
49:58third of the time
50:00and at, you know, a
50:01third of the cost.
50:03You're probably also getting a
50:04lot of other different contrast
50:06that you're maybe not used
50:08to looking at or aren't
50:09as useful,
50:10but there's been tremendous development.
50:12Just because you're still looking
50:13at fat water doesn't mean
50:15that there's not tremendous development.
50:17And so I think that
50:18applies to a lot of
50:19the stuff you guys are
50:20discussing there.
50:31Other questions?
50:35So so I heard
50:37John and Eric, you both
50:38said oh, I'm sorry.
50:40So one sec.
50:42That, you know, we're not
50:43using enough imaging.
50:46What we've heard for the
50:46longest time was that, well,
50:48imaging is too expensive. Can
50:49I just say, I'm not
50:51sure that it's not that
50:51we're not using enough? I
50:53think we're not using it
50:54in the right way. Yes.
50:55Yes. I mean, I think
50:56we're using too much of
50:58standard imaging.
50:59And I think, you know,
51:00we could learn to back
51:01off on that and then
51:03add in
51:05other So so I wanna
51:06take one example about this
51:07where I think it's not
51:08true,
51:09and that's when we're looking
51:10at therapies
51:11because the therapies we're looking
51:13at are incredibly expensive.
51:15And if imaging can save
51:16you, in theragnostics,
51:18one more treatment,
51:20that's a hundred thousand dollars,
51:22and a scan is
51:23a thousand dollars, which is
51:25fifteen hundred, right, give or
51:26take. And we do everybody
51:28six six of those cycles
51:30for everybody, but that's a
51:32huge chance.
51:33But we don't. Actually, in
51:34in the clinic, we don't,
51:36do personalized treatment. We just
51:38do
51:39you go you do your
51:39six cycles and you come
51:40back. And I think in
51:41AD, it's the same. Like,
51:43if we could in groups
51:44of that are more at
51:45risk early on, we can
51:47see, well, whose
51:48tau is going down, who's
51:49not, it's not one hundred
51:51thousand dollars but it's fifty
51:51thousand dollars for the kind
51:52of math, and that's also
51:53a statement. Why aren't we
51:54doing it?
51:58You know,
52:00I feel there's a balance
52:01here, sort of like the
52:02comment that was made up
52:03there. I do think we've
52:04made a lot of progress
52:05in treatment selection.
52:08Being an acute stroke doctor,
52:10we've just had a complete
52:11change, right, in our paradigms
52:13in how we imaging select
52:15patients to open up their
52:17large artery occlusions using CT
52:19perfusion.
52:20Or if you look at
52:20Alzheimer's disease that, you know,
52:22we're looking for those patients
52:24with amyloid plaque burden that
52:25we're giving lecanumab
52:26to. So
52:28I do think you know,
52:28there's been a ton of
52:29progress to, to be excited
52:31about and more to be
52:32had, but I think where
52:33I, I, I also agree
52:35with Eric is there some
52:36opportunities to fine tune, make
52:38it more cost effective,
52:39make it just more intelligent
52:41in our clinical approaches. And
52:42some of that isn't as
52:44exciting as sort of the
52:45highest tech things we do,
52:46but they're equally important.
52:49Yeah. I would I would
52:50say Oh, there we go.
52:51I I would say number
52:52one to to your point.
52:53I mean, imaging has gone
52:54a lot has come a
52:55long way. I remember I
52:56remember reading CT scans that
52:58were, like, five millimeter thick
53:00slices,
53:01and you would have ten.
53:02And now you have point
53:03nine millimeter slices, and you
53:05have hundreds, and you can
53:07see with much greater resolution,
53:08and you can collect those
53:09data much faster than you
53:10could even those fifteen thick
53:11slices. So I think we've
53:13come a long way. But
53:14to to your point, George,
53:15that,
53:16you know, there there's this
53:17opportunity and and to what
53:19Eric was saying to maybe,
53:21get a little bit more
53:22personal and leverage imaging to
53:24choose the right
53:26therapy at the right time
53:28for the right number of
53:29doses. And even and we're
53:31just at the precipice of
53:32getting even more specific
53:34to look at thing you
53:35know, tumors that are heterogene
53:36heterogeneous. And you can imagine
53:38you can envision one day
53:39that there could be specific
53:41diagnostic
53:42cocktails of
53:43different that, that target different,
53:46heterogeneous targets within a tumor
53:48and treat them incredibly efficiently
53:50for that individual,
53:51which could limit, you know,
53:52rather than a one size
53:53fits all,
53:55which obviously is just the
53:56hammer, one size fits all,
53:57and and that ultimately it
53:59needs to get more personal.
54:00And that's when you can
54:01sort of get a little
54:02bit more cost effective in
54:04the way you do things
54:05when it's very personalized,
54:06and you and then you
54:07can get more selective
54:09in choosing the the patient
54:11who's gonna respond best so
54:12that you're not just treating
54:13everybody with the same cost.
54:15You're really honing in on
54:16the group that's gonna respond
54:18best and, therefore,
54:19avoiding unnecessary cost and unnecessary
54:21toxicity when when someone doesn't
54:23need it. So So, I
54:24mean, we've made tremendous progress.
54:26There's no question. But so,
54:27you know, the study reported
54:29three days ago, you know,
54:31fourteen
54:32highly toxic treatments were better
54:34than fourteen
54:36less toxic treatments
54:37for in a specific situation.
54:39But wouldn't it be better
54:41if through the use of
54:42imaging, we could figure out
54:44that,
54:45you know, after that in
54:47eighty percent of people, after
54:49four treatments, everything seemed gone?
54:51And then,
54:53you know, and and then
54:54you could then confirm that
54:55with a subsequent study and
54:57just
54:58limit it to four treatments.
54:59Absolutely. Or you could you
55:00could choose the subset that
55:02needs six and the subset
55:03that needs four. And then
55:05now that you've you've become
55:06more efficient and more cost
55:07effective. Alright. So we have
55:09two questions. Sylvia, you've been
55:10very patient. Go. Okay. Is
55:12this on? Okay. I have
55:14a question and then maybe,
55:15like, a background comment for
55:17radiology.
55:18So is there, like, what's
55:19on horizon for, like, a
55:21straightforward
55:22time effective volumetric
55:24MRI assessment?
55:26It's repeat that again. A
55:27single Volumetric MRI.
55:30Oh. It's, like, time consuming.
55:31Right? Oh. Oh, yes. Yes.
55:33So I think instead of,
55:34like, a PI assisted, like,
55:35you know, what's there to
55:36come. Yeah. Yeah. For a
55:37background, I want to kind
55:38of say, you know, I'm
55:39a neuro oncologist, so we
55:40do meningiomas.
55:43So I'm actually working on
55:44the manuscript
55:45about theranostics in meningiomas.
55:48And I already know, like,
55:49the reviewers will kind of
55:51shoot me down for not
55:52having volumetric MRI
55:53assessment
55:54of my tumors. Right. Because
55:56our rhino assessment, which is
55:58like the neuro resist criteria,
55:59right, still revolve around t
56:02one, t two, post t
56:03one,
56:05gadolinium
56:06enhanced Right. Energy. Well, we're
56:08we're I will say, like,
56:08the two last iteration of
56:10the renal criteria also talked
56:11about volumetric assessment and how
56:13does this the future, but
56:14it's not available. Yeah. Well,
56:16the good news is it's
56:17it's it's here, and it's
56:19slowly being implemented. And there's
56:20two ways that this is
56:21rolling out. First of all,
56:23MRI scanners, you know, to
56:25to to our point, have
56:26gotten much, much, much better.
56:27And so the and just
56:28for the audience,
56:30you know, what we're talking
56:31about is volumetric,
56:33and iso you know, isotropic
56:34volumetric image. So,
56:36because because MRI takes a
56:38long time to acquire, historically,
56:40what we've done, let's say,
56:41for the brain is we'll
56:42acquire a slice at, like,
56:43I don't know, one millimeter
56:45slice, skip a millimeter. Another
56:47and then acquire another one
56:48millimeter slice, skip one. Acquire
56:50one, skip one, all through
56:51the brain. So even when
56:53you're scrolling, it looks like
56:54you're looking at the whole
56:55brain, but, really, you've only
56:56imaged half the brain. And
56:57so it's not truly volumetric,
57:00which does can interfere with
57:02making accurate measurements of things
57:04like meningiomas.
57:06And the reason that is
57:07is because it takes a
57:07long time to acquire the
57:08data. And if you're if
57:09it takes, let's say, twenty
57:11minutes to acquire a volumetric
57:12dataset, and that's not an
57:14accurate number, but let's say
57:15twenty minutes.
57:16Number one, it means that
57:17you're creating a backlog and
57:19it limits access. And number
57:20two, patients start to fidget
57:22around when they're in an
57:23MRI scanner for a long
57:24time. And so if you
57:25if you if you acquire
57:26a really long acquisition and
57:27you get a volumetric image
57:29only for them to start
57:30moving around, it ruins your
57:31data and then you have
57:32to repeat it and then
57:33that's another twenty minutes. But,
57:34with AI acceleration, now we're
57:36getting to the point where
57:37we can acquire volumetric images
57:39really, really fast.
57:41I'd say that there there's
57:43still some demand because of
57:44the because of the need
57:46for access and to get
57:47patients through. There's still a
57:48lot of backlog.
57:50So even though we can
57:51acquire a volumetric image a
57:52lot faster,
57:54for if if you're if
57:55the interpretation is what I
57:56described before, which is qualitative,
57:58you don't really need to
57:59see every slice to make
58:00a qualitative interpretation of an
58:02exam
58:03to then refer the patient
58:05or or or get a,
58:07have a a helpful response
58:09to your oncologist to treat
58:10the patient. But it does
58:11interfere with that quantitative part
58:13that you're talking about, which
58:14is the future of the
58:15field having, you know, really
58:16using leveraging quantitative data to
58:19better treat the patient. That
58:20that day is coming, and
58:21it does require that kind
58:22of volumetric dataset.
58:24So so one way that's
58:25happening is, again, through AI
58:26acceleration, and now we can
58:28acquire a whole volumetric dataset
58:30in just a few minutes.
58:31But but now well, it's,
58:33let's say, three minutes. But
58:34if you can acquire the
58:35same kind of clinical dataset
58:37that I just talked about
58:37in one point five
58:39and that increases access, then
58:40you're always kind of balancing
58:42patient access and the need
58:43to push patients through with,
58:46you know, quality of data.
58:47And what what is the,
58:48quality of data that you
58:49need to make the diagnosis?
58:51And it's probably that lower
58:52version where you don't need
58:53the whole brain,
58:55but it that's not that's
58:56not what pharmaceutical
58:57companies want when they're doing
58:59a clinical trial. They want
59:00the volumetric data. So one
59:01way is through AI acceleration.
59:03There's also other ways to
59:04take a semi quantitative
59:06nonastropic data set and then
59:08use generative AI
59:10to then fill in the
59:11gaps, and that can also
59:13work reasonably well. You're still
59:15sort of interpolating,
59:18data that's not there. But,
59:20that is one approach that
59:21appears to be
59:23having some
59:24positive yield in
59:27getting these more quantitative measurements
59:29that you're talking about. So
59:30those are so they're here,
59:32and now they're available to
59:34roll out, but I think,
59:35you know, this is where
59:36collaboration comes into play. And,
59:38and there's no reason well,
59:40there are lots of reasons
59:41that why this might be
59:42complicated, but you could you
59:43could, let's say, on a
59:44clinical service, you could have
59:46your standard
59:47brain tumor imaging protocol, and
59:48you could have your advanced
59:50brain tumor imaging protocol. And
59:51if it's for a clinical
59:52trial, maybe that's the one
59:53that has the,
59:55isotropic fully quantitative
59:57volumetric
59:58t one dataset.
60:00Maybe that's the one that
01:00:00has a three d flare
01:00:02instead of a two d
01:00:02flare. Maybe that's the one
01:00:04where you even include some
01:00:05ASL,
01:00:06blood flow imaging and maybe
01:00:07some other advanced,
01:00:09measurements that that, our physicians
01:00:11can start to leverage and
01:00:12while we're taking you know,
01:00:13to help take care of
01:00:14patients better. And are these,
01:00:15like, standard, I think, that
01:00:16you can kinda roll out
01:00:17to various So, Sylvia, I
01:00:19I don't want us to
01:00:20cut into Dean Brown's closing
01:00:21comments.
01:00:22And there was one last
01:00:24quick question maybe, Larry. Oh,
01:00:25just a quick comment. And
01:00:26we will be able to
01:00:27continue this discussion because we
01:00:28have an hour and a
01:00:29half of poster afterwards. Just
01:00:31a quick comment. Thank you
01:00:32for a wonderful day and
01:00:33thank you for putting this
01:00:33together, Dean Brown and George
01:00:35and everyone. I just wanted
01:00:36to echo what, doctor Whitlow
01:00:37and doctor Weiner were saying
01:00:39about the importance of tumor
01:00:40heterogeneity.
01:00:41And that's exactly what we're
01:00:42focusing on with PETCT,
01:00:44whether we're doing multiplex
01:00:46PET with FDG,
01:00:47PSMA or DOTATATE. So tumor
01:00:50heterogeneity is something that we
01:00:51see every day in PET.
01:00:53And ideally, just to comment
01:00:54about theranostics, it's unusual for
01:00:56patients to get six treatments
01:00:57for, a Pluvicta and the
01:00:59goal is to personalize their
01:01:01ownostics exactly as you're saying
01:01:03with dosimetry, and that's what
01:01:04George is helping us with.
01:01:05So that's our goal, personalized
01:01:07their ownostics. So thanks. Thank
01:01:09you. So,
01:01:11we're almost done. We have
01:01:12our closing remarks from Dean
01:01:14Brown, and then we have
01:01:14a poster session. We have
01:01:16forty two posters outside.
01:01:18There will be food and
01:01:20refreshment,
01:01:21at different levels. There are
01:01:23brain posters,
01:01:25cardiac,
01:01:25cancer, and other areas.
01:01:27Please do go and see
01:01:28some of them. Please ask
01:01:30more questions if you'd like,
01:01:31and I'll give the floor
01:01:32to Dean Brown. Thank and
01:01:34please scan this QR at
01:01:35the end,
01:01:37to tell us how we
01:01:38did, whether bad or good.
01:01:39That would be useful for
01:01:40other
01:01:42dean's workshops and for other
01:01:43meetings we would do.
01:01:45Great.
01:01:46That that's like my whole
01:01:48role in the final remarks,
01:01:49by the way.
01:01:51This has been really this
01:01:52has been really great. I
01:01:53can't wait to hear the
01:01:54conversation
01:01:55after you
01:01:56have some of the drinks
01:01:57that are out there.
01:02:00I I heard we need
01:02:02more, which I took to
01:02:03mean money.
01:02:05And then I got really
01:02:06nervous about the whole, like,
01:02:07stick Nancy in the MRI.
01:02:08I thought we were getting
01:02:09into, like white matter disease
01:02:11or something. So, anyway, really
01:02:13great day and thanks to
01:02:15Ruth and Beth and the
01:02:16team and thanks to all
01:02:18of you who spoke and
01:02:19put it on.