Pathology Grand Rounds Feb. 12, 2026 - Liron Pantanowitz, MD, PhD, MHA

Name: Pathology Grand Rounds Feb. 12, 2026 - Liron Pantanowitz, MD, PhD, MHA
Uploaded: 2026-02-20T16:02:21.6566667Z
Duration: 1 h 40 min 3 s
Description: Pathology Grand Rounds Feb. 12, 2026 - Liron Pantanowitz, MD, PhD, MHA, Maud Menton Professor and Chair, Department of Pathology, University of Pittsburgh, University of Pittsburgh Medical Center, on, "The Digital Pathology Journal at UPMC."

February 20, 2026

Pathology Grand Rounds Feb. 12, 2026 - Liron Pantanowitz, MD, PhD, MHA, Maud Menton Professor and Chair, Department of Pathology, University of Pittsburgh, University of Pittsburgh Medical Center, on, "The Digital Pathology Journal at UPMC."

Information

ID: 13856
To Cite: DCA Citation Guide

Download Transcript

00:00Is my great pleasure,
00:02to introduce our today's grand
00:04round speaker, doctor
00:06Liron Pantanovic.
00:07Actually, Liron doesn't need any
00:09introduction, but I'm going to
00:10do it anyway.
00:11So,
00:14Liron is I'll just start.
00:16He's the godfather of digital
00:17pathology. He's the godfather of
00:19AI in pathology, and it's
00:21a real blast to have
00:22him here. Thank you so
00:23much for coming, Liron. And,
00:25in terms of, like,
00:28you know, his education and
00:29everything, Liron actually is,
00:31is started his undergraduate and
00:33graduate studies
00:35in South Africa where he
00:36actually initially also practiced a
00:38little bit. Then, he moved
00:40to US in nineteen ninety
00:41nine where he did his
00:43residency
00:44in APN CP at Tibet
00:46Israel, so in the neighborhood
00:48in up in Boston.
00:50He, also did their hematopathology
00:53fellowship
00:54and followed by the cytopathology
00:56fellowship in a Bay State,
00:57medical center. As a pathologist,
00:59he started his career first
01:01as an instructor in a
01:02Harvard system,
01:04and then,
01:05he moved to Tufts in
01:06two thousand five, stayed there
01:08for five years,
01:10and then came to Pittsburgh
01:12in two thousand
01:13two thousand ten. I couldn't
01:14believe it. It was two
01:15thousand ten. So it was
01:17really my great privilege to
01:19know Liron at that time
01:20and,
01:22he's absolutely brilliant pathologist. He's
01:26definitely independent thinker,
01:28progressive thinker.
01:31He
01:32published
01:33over you stopped your CV
01:34in twenty twenty three, over
01:35six hundred
01:37manuscripts, book chapters, everything else,
01:40numerous talks across the world,
01:42and he's really the one
01:44who actually paid the way
01:45for the digital pathologist to
01:47digital pathology. And he had
01:49idea of how digital pathology
01:51should work in a clinical
01:53a clinical sense twenty twenty
01:55five years ago, and it's
01:57actually great to see that
01:58this is coming now in
01:59life. So I'll stop there,
02:01Lerim, because I can go
02:02forever about you, but thank
02:04you so much. Thank you.
02:12So good afternoon, everyone.
02:15And thank you, doctor Dacek,
02:16and thank you for inviting
02:17me to be your grand
02:18round speaker.
02:20So in October of two
02:21thousand and thirteen,
02:24I was standing right here.
02:26So doctor Morrow had invited
02:28me for grand rounds back
02:29in two thousand thirteen. And
02:31guess what I was asked
02:32to talk about? Digital pathology
02:34at UPMC.
02:35So I'm hoping in the
02:36twelve plus years that I
02:37can update,
02:39whatever I spoke to you
02:40about then. This is not
02:41a just a rehash of
02:42the same old PowerPoint.
02:46Okay. That being said,
02:48I do wanna let you
02:49know,
02:50I work a lot with
02:51industry either on advisory boards,
02:53consult with them. I have
02:54these two startups which have
02:55gone nowhere. In fact, they've
02:57only cost me money.
03:00But I wanted to disclose
03:02that, and none of that
03:02should really influence the contents
03:04of this talk.
03:06So my objective is pretty
03:08simple.
03:09There are three of them.
03:10One,
03:11I wanted to review with
03:12you the evolution of digital
03:13pathology at UPMC,
03:15and you'll see it was
03:16not an easy pathway. And
03:17we are still on that
03:19evolutionary
03:20pathway because
03:21when when people say we're
03:22going digital or we're going
03:23to start digital,
03:25what the one thing they
03:25don't realize is it's a
03:27never ending pathway. It's like
03:28when you buy your first
03:29iPhone,
03:30you know, next year there'll
03:31be another version of an
03:32iPhone, and you're going to
03:33have to learn to use
03:34it and so forth.
03:36I do want to point
03:37out that along this journey,
03:39at UPMC, none of that
03:40was easy. None of it
03:41was perfect. We made a
03:42lot of mistakes. And so
03:43I know that you're on
03:44a digital pathology journey of
03:46your own. And so I'll
03:47share some of our lessons,
03:48and hopefully
03:50they'll benefit you.
03:51And finally,
03:53everything we did, every camera
03:55we purchased, ultimately helped build
03:57the infrastructure,
03:59helped get people more savvy,
04:01and receptive to using technology,
04:03and our own institution to
04:05understanding that we need to
04:06evolve and be flexible in
04:08order to keep up with
04:09the practice of medicine so
04:10that we can have an
04:11infrastructure in which we can
04:12actually use artificial intelligence.
04:17So my belief simply
04:19with digitizing slides, and I'll
04:21confine most of my talk
04:22to anatomical pathology,
04:25and a lot of that
04:26to,
04:27surgical pathology
04:28because there's a lot that
04:29can be said about lab
04:31medicine
04:32going digital too as well
04:33as cytology,
04:35is is the minute you
04:36digitize your glass slide,
04:38you gain the portability
04:40and the applications you can
04:41run on that. And I've
04:43had many conversations
04:45with some very serious famous
04:46pathologists,
04:48and I've told them simply
04:49that, who have were very
04:51resistant to going digital,
04:54that the reason they have
04:56their own mobile phone
04:58is, one, that they don't
05:00have to be in their
05:01office to use their landline
05:02to make a phone call
05:03even though they know that
05:05the phone call you'll make
05:05in your office is way
05:07more clear than using your
05:08mobile phone, traveling, going through
05:10a tunnel, but you you
05:12still enjoy the portability.
05:14But more than that is
05:16all the apps you have
05:17on you. Check your email,
05:19use your GPS, now chat,
05:20GPT, etcetera. And that would
05:22not be possible
05:24if you don't digitize your
05:25slides.
05:27So, as a chair, there
05:28are several things that I
05:30would like to accomplish in
05:31my department,
05:32and I think I've already
05:33shared these with,
05:35your chair. But the first
05:37is medicine is a very
05:38connected field. And, you know,
05:40unless we're going to stay
05:41in the basement with our
05:42paperwork and microscopes forever, no
05:44one will know us.
05:45So I am trying to
05:47consolidate everything where I am.
05:49We're a forty plus hospital
05:50system.
05:52Pathologists are not all in
05:53the same area. So if
05:55I can consolidate that, in
05:56fact, I have to consolidate
05:57that. I just don't have
05:58enough cytotechnologists
06:00and histotechnologists
06:02to put in labs
06:03all over the place.
06:05We definitely
06:06need to need to be
06:06efficient.
06:07Someone's working from home. Someone
06:09needs coverage.
06:11The only way I can
06:11really do that is with
06:12telepathology.
06:13The only way to reduce
06:15turnaround time, majorly between hospitals,
06:18is by using digital images.
06:21And for me to have
06:21someone pull an old slide,
06:23if someone has a resection
06:24or they're going to do
06:25a frozen section, they would
06:26love to see a prior
06:27immediately within a minute, not
06:29put in a request
06:30that may never actually deliver
06:32a slide to you is
06:33to go digital.
06:35Plus the distribution of the
06:36workload. So we, in Pittsburgh,
06:38have a very subspecialized
06:40practice, a lot of, you
06:41know, people in different areas.
06:43And to support the subspecialization,
06:45they need the cases sent
06:46to them. And throughout our
06:48system, the best way to
06:49refer cases to them is
06:51from community to do that
06:52digitally these days.
06:54And if you don't go
06:55digital and you don't integrate,
06:57you can forget about using
06:59AI effectively.
07:02Now the hospital and the
07:03dean obviously want to know
07:04from me, well, Iran, if
07:05we give you all this
07:06money to invest in technology,
07:09where's the revenue? Where can
07:10you make money out of
07:12using the technology? Or how
07:13are you going to save
07:14costs? And, you know, that
07:15is a difficult question,
07:18to answer. But also importantly
07:20is where's the quality that
07:21you add? Is there a
07:22value to doing this?
07:24And, of course, lower liability.
07:27It just takes one twenty
07:28million dollars lawsuit that could
07:30have been prevented using technology.
07:32We learned that mistake back
07:34in the barcoding era. There
07:35was no money to be
07:36made in barcoding all your
07:38slides and your blocks. But,
07:39if there was a mix
07:40up of a case,
07:42if you didn't barcode,
07:44you know, that was well
07:45worth the money spent.
07:47We are in this era
07:48of computational pathology, which is
07:50AI.
07:52And I think it's essential
07:54for pathologists and trainees to
07:55know how to create, validate,
07:57and use AI.
07:59Most people don't even know
08:00how to monitor AI.
08:02And there are lots of
08:03companies who want to work
08:04with
08:06industry and academia to join
08:08that partnership to advance technology.
08:10They cannot do it on
08:11their own. And frankly, the
08:13way that federal funding and
08:14everything else is going, it
08:16makes a whole lot of
08:17sense to work with industry
08:18to try and get stuff
08:19done.
08:21Plus, the residents that apply
08:23to our programs now, the
08:25fellows that want to come
08:26work at UPMC, and even
08:27the junior faculty who
08:30went to residency programs that
08:31were fully digital do not
08:32want to take jobs with
08:33us unless we're fully digital.
08:35And so there are many
08:37residents that request to speak
08:38to me because they want
08:39to know my digital pathology
08:41roadmap before they match and
08:43rank us, believe it or
08:44not.
08:46And, of course, the reputation
08:47of Pitt.
08:49We want to be known
08:49as a place in Pittsburgh
08:50where we do cutting edge
08:51technology. We do not wanna
08:53be known as
08:54a steel industry city like
08:56we were fifty years ago.
08:58Two, we want our clinicians
09:00and our patients to know
09:01that we can offer them
09:03the latest and greatest cancer
09:05diagnostics,
09:06prognostics, etcetera.
09:07So the next in vitro
09:09diagnostic test that comes out,
09:10and there are many that
09:11pharma are working on, if
09:12you can only do that
09:14by scoring it with a
09:16computer
09:17because look at the mess
09:18with PDL one, right, how
09:20How we have to score
09:21that manually and all the
09:22different scoring systems. We wanna
09:23be at the forefront of
09:24that.
09:26And people wanna come and
09:27work with us. We opened
09:28an AI center just over
09:30a year ago,
09:31and I get about
09:33twenty
09:34emails
09:35every month from people wanting
09:36to come and work
09:38in our AI center with
09:39us, etcetera. So we wanna
09:40attract talent. Now here's the
09:42reality, okay, which many of
09:44you know. I'm not yet
09:45to sell you your vacation
09:47on the left, which is
09:48what the postcard and the
09:49flyer does, okay, because that's
09:51what the digital pathology vendors
09:53tell you.
09:55Buy the system and install
09:56it, and that's what life
09:57will be like. Reality is
09:58what it more like on
09:59the right. When you arrive
10:00at your
10:01paradise beach vacation,
10:04that's what digital pathology looks
10:05like. Okay. It's messy and
10:07it's difficult and, you know,
10:08the IT systems aren't integrated
10:10and things are more expensive
10:11than you had anticipated. So
10:13let me actually
10:14give you a reality check
10:15and not paint paradise for
10:17you. So just a little
10:18bit of what are we
10:18talking about when one talks
10:20about digital pathology workflow.
10:22So the most important thing
10:23about digital pathology, which hopefully
10:25you'll get throughout this talk,
10:27is the pre imaging part.
10:28Before a slide even gets
10:29onto a scanner,
10:31if you haven't done certain
10:32things, your digital pathology will
10:34fail. But the pre imaging
10:36side starts with scanning a
10:38slide, and if you haven't
10:39barcoded that slide, you can
10:40forget about an automated system
10:42that works well. And if
10:43you don't have the perfect
10:45barcode, believe it or not,
10:47about ten
10:48percent or more of those
10:49slides will get scanned, but
10:50they're going to cyberspace. They're
10:52not going to get linked
10:53to your LIS or wherever
10:54you want. Okay. So you
10:55really need good barcoding,
10:57practices in place.
10:59And then on the top,
11:00as you can see, how
11:01you're going to also scan
11:03cytology, microbiology,
11:05hematopathology,
11:06and fluorescence images. For those,
11:08you kind of need some
11:10niche workflows and niche,
11:12scanners. But once you've scanned
11:14your slide,
11:16you can store those
11:17and ideally link it to
11:19your lab information system. I've
11:20always been a proponent of
11:22practicing pathology with an LIS
11:24driven workflow. Not that I
11:26love the lab information system,
11:27but that's where
11:29our workflow goes through. That's
11:31where we do reporting and
11:32auditing.
11:33That's,
11:34you know, where billing happens.
11:36And wherever billing happens, that's
11:38where they want you to
11:38work.
11:40You can store it with
11:41an enterprise. The The VNA
11:43stands for a vendor neutral
11:44archive. So if your institution
11:46has an imaging strategy that
11:48all images, radiology, pathology, all
11:50other ologies will be stored
11:51somewhere and either viewed
11:54by everyone in practice or
11:55at least stored and used
11:57for research, you can do
11:58that.
11:59And then, of course, when
12:00the image is available for
12:01someone on a workstation,
12:03all the applications are endless.
12:05You can share it remotely.
12:06That's telepathology.
12:08You can make a diagnosis
12:09of that image, which is
12:11primary diagnosis, meaning you don't
12:12need the glass slide anymore.
12:14You can analyze the pixels,
12:16quantitative image analysis or AI,
12:19and then use it for
12:20research, education, and so forth.
12:22Okay. So this is the
12:23evolution of how digital pathology
12:25unfolded at UPMC in this
12:27actual order.
12:29One was virtual education, two
12:32tumor boards, frozen section telepathology,
12:35then teleconsultation.
12:37Then we added telecytology for
12:39the rapid on-site e dials
12:40in cytology.
12:42Then we went to primary
12:43diagnosis.
12:44Then we opened an image
12:45analysis lab to do quantitative
12:46image analysis.
12:48And now we're using AI.
12:49So I'll run through that
12:50with you. The pictures I've
12:52shown you are actual real
12:53pictures from UPMC. We've been
12:55through every kind of hardware,
12:56software, camera, scanner,
12:58homegrown,
12:59commercial,
13:00and so we've learned from
13:02that. And not only have
13:03we learned from that, but
13:04our IT department has learned
13:06from that. We have IT
13:07analysts that have used all
13:09of these instruments and are
13:10very comfortable, you know, with
13:12modern technology having gone through
13:14this process with us.
13:16So the first thing we
13:17did was virtual education. Our
13:19residents back in the day
13:20wanted to look at slides
13:21virtually. UPMC was unfortunately very
13:24geographically dispersed. We had hospitals
13:26all over Pittsburgh. Now residents
13:28had to rotate and how
13:29are they gonna do unknown
13:30conferences and so on. So
13:32they wanted a virtual
13:35platform to do that. There
13:36was nothing commercially available. So
13:38and Neil Parwani and myself
13:39back in the day working
13:40with IT, we built our
13:41own homegrown system.
13:43I'll tell you that it's
13:44better than any commercial system
13:45out there today. Nevertheless,
13:47UPMC didn't want us to
13:49build. They wanted us to
13:51buy. So we bought PathXL,
13:53worked wonderfully. Then Philips bought
13:54it and shut it down.
13:56And now we use Path
13:57Presenter.
13:58But the picture I've shown
13:59you there is a resident
14:02who wanted to do a
14:03one month elective with me
14:04in Cytology.
14:05And that, by the way,
14:06is probably fifteen years ago.
14:11You will notice, and I
14:12took a picture as I
14:12was walking by, she has
14:14pushed the microscope off to
14:15the side and she's looking
14:17at all the cytology teaching
14:18sets on a monitor. And
14:20boy, that's a terrible monitor,
14:21such low resolution. But she
14:23spent the month going through
14:24all our teaching sets because
14:26she had a natural affinity
14:27to doing it digitally.
14:28Maybe that's why she became
14:30a dermatopathologist
14:31today. But, nevertheless,
14:33that was our first foray
14:34because the residents forced us
14:36to go digital,
14:37and they still do today.
14:40The next was
14:42our residents and fellows got
14:43tired of running tumor boards
14:45the old fashioned way, and
14:46they were the ones that
14:47pushed us.
14:49That is Soumyk Roy.
14:51He was a a resident
14:52in our program. This is
14:53his year when he did
14:54his GU fellowship. Then he
14:55became a molecular pathology,
14:58guru, and now he's a
14:59big deal at the University
15:00of Cincinnati.
15:02But what Soumyk did was
15:03Soumyk was, you know, very
15:05digitally savvy, liked to code
15:06for a hobby on the
15:07side, and he came to
15:08me and said, Doctor. Bentano,
15:09it's
15:10the worst part of my
15:11GU fellowship is every other
15:13day I've got to go
15:14and find the slides. I've
15:15got to take photos and
15:16put them in a PowerPoint
15:17presentation.
15:18And then I have to
15:19show up at the tumor
15:20boards.
15:22And every time, you know,
15:23if it's a different room,
15:24I've got to figure out
15:25how to plug in this
15:25USB, and and, it's asking
15:27me to reformat the drive.
15:29And then the GU
15:31surgeons, the urologists are asking
15:33me questions that had nothing
15:34to do with the pictures
15:35I took, and I couldn't
15:36show them.
15:37So it was embarrassing. Whereas
15:39the radiologist would just show
15:40up
15:41one minute before the a
15:42coffee in hand, log on,
15:43and they could see anything.
15:44Go back. Look at the
15:45CT, that CT. Like so
15:46can can we go digital?
15:48And so I said to
15:49Soren, let's do it.
15:51And so he used his
15:52own iPad. All the GU
15:54slides got scanned,
15:56and he was fully prepared
15:57for TumorBoard.
15:59So today,
16:00and I did this recent
16:01calculation because UPMC asked me,
16:03where's the return on investment
16:04if we digitize all forty
16:06of your hospitals?
16:08So we run fifty TumorBoard
16:10throughout our UPMC system.
16:13Many hours are invested in
16:14that. And we calculated that's
16:16three FTEs a week to
16:17run those tumor boards.
16:20Now that most of those
16:21are digital, not all, but
16:22most of those are digital,
16:23I'm saving three FTEs worth
16:25of time and obviously administrators
16:28like that. So
16:31the next advantage of going
16:32digital is that,
16:34you can run your tumor
16:35boards. And whether you scan
16:36prospectively
16:37or retrospectively, you still can
16:39run your tumor boards.
16:41Now, the number one application
16:43for me is if you
16:44go digital is that you
16:45can share that image.
16:47And I still think that's
16:48the number one application even
16:49more so than AI.
16:51It's way easier to move
16:52an image than to tell
16:53a patient You can only
16:55get your neurosurgery done at
16:56this facility because that's where
16:58we can do the frozen
16:59sections,
17:01or a particular procedure. And
17:03it's so much easier and
17:04actually less expensive to move
17:06an image around them than
17:08to have a pathologist move
17:09around, to go do a
17:10frozen or to go look
17:11at a cytology touch grip
17:12or imprint.
17:14So in the acute setting,
17:15you can do frozen sections
17:17or wrap it on-site emails
17:19for cytology digitally. Makes a
17:21whole lot of sense.
17:23If you have time,
17:24you can send it to
17:25someone to make a primary
17:26diagnosis,
17:27or you can receive a
17:28consult. So that's
17:30the advantage.
17:32I often get calls,
17:34even from Yale,
17:36Like, which system should we
17:37use? And that's for me,
17:38like, which system are we
17:39buying? And the problem is
17:40we have so many systems
17:41now that you can pick
17:43live views, whole site imaging
17:45views, etcetera.
17:46And I simplify it as
17:47this. So if you're attaching
17:49something to your camera mean,
17:51sorry, if you're attaching something
17:51to your microscope or camera,
17:53whether you're streaming a video
17:54or taking a picture,
17:56you need someone on the
17:57other end who knows what
17:58they're looking at, such a
17:59technologist,
18:00a pathologist, etcetera,
18:03in order to practice telepathology.
18:06But if you can scan
18:07the whole slide without having
18:08anyone skilled, just a PA,
18:11maybe a resident in their
18:12first year, or even someone
18:14who's like a radiology,
18:17assistant
18:18who can just load the
18:19slide for you,
18:21then you can use robotic
18:22or whole side imaging.
18:24And and we've used and
18:25are using all of these
18:26different setups.
18:29The first time we decided
18:30to do frozen sections remotely
18:31was,
18:33in two thousand and one,
18:35which is a long time
18:36ago.
18:37And
18:38at some point, we had
18:40no option
18:41because
18:42we were covering frozen sections
18:45for transplants
18:46in Italy.
18:48Okay? So, our hospital was
18:50supporting,
18:52a hospital in Italy to
18:53do bring up transplant and
18:54do frozen sections. And they
18:56didn't know even how to
18:57interpret frozen sections for transplants.
19:00We agreed to do that.
19:01And so, we did it
19:02by telepathology.
19:03We're very dedicated transplant pathologists.
19:05Dedicated, but difficult. Right, Sonya?
19:07Yes. Okay. So, but they
19:09would wake up at two
19:10in the morning and do
19:11frozen.
19:12And, you know, through that
19:13we learned. And then our
19:14neuropathology
19:14team followed suit, too.
19:19So today,
19:20we have these LV1
19:23systems from Leica.
19:25They're all around many hospitals.
19:28Some are even located, you
19:29will see, hundreds of miles
19:31away where we do
19:33tons of frozen sections a
19:34day
19:35remotely.
19:36We were just being told
19:38by Leica that they're no
19:39longer going to support the
19:40LV1, so we're looking at
19:42new options to do frozen
19:43sections,
19:44digitally.
19:46But what had happened over
19:47time is concordance got better.
19:49So in the beginning when
19:51we were doing this ten,
19:52twenty years ago,
19:54our pathologists made mistakes. And
19:56so concordance to the
19:58permanent,
19:59afterwards, now is way better.
20:02Two, we don't
20:03defer that many cases anymore.
20:05In the beginning, they weren't
20:06sure, so they would defer
20:07to permanent. Now they hardly
20:09defer cases.
20:10But this is Pandora's box,
20:12which is a downside. The
20:13service got way busier.
20:15Now that, for example, our
20:16neurosurgeons know they can operate
20:18in any hospital
20:20and just like going to
20:21a vending machine and pushing
20:23and saying, I want a
20:23soda right now, they can
20:25do that. And so our
20:27neuropathologists
20:28are constantly barraged by frozen
20:30sections
20:31all the time every day.
20:32They basically have to have
20:34a neuropathologist
20:34uncovering frozen
20:36twenty four seven.
20:37So that is a downside,
20:39that, you know, and we've
20:40enabled.
20:42So this is, one hospital
20:43today. You can see Shadyside
20:45Hospital in this particular frozen
20:47section room just so you
20:47can see where we evolved
20:49back in the day by
20:50someone taking a picture in
20:51Italy and sending it to
20:52pathologists in Pittsburgh. Now we
20:54have two of these instruments
20:56in the frozen section room.
20:57One is for neuropathology because
20:59it's so busy.
21:00One is for all the
21:01other surgical pathology systems. Everyone's
21:04trained. We can have a
21:04first year resident who knows
21:06what to do after hours,
21:08validated, and it really works
21:10well and it's streamlined. And
21:11it's become common practice.
21:13No one sort of
21:14questions that, well, how are
21:16we doing it? Or, wow,
21:16this is modern technology.
21:18It's as like, we wouldn't
21:19have a frozen section room
21:22without a microtome. We wouldn't
21:23have a frozen section without
21:24the ability to telepathology.
21:29But what we did learn
21:31and what did happen is
21:32our surgeons missed us, believe
21:34it or not.
21:35I was just we didn't
21:36miss them, but
21:39but they missed us being
21:40there in the frozen section
21:41room.
21:43They liked to walk in
21:44there,
21:45chat, harass us,
21:47question the diagnosis, or whatever.
21:49They liked us being there
21:50physically present, and now they
21:52were not.
21:53So we were not there.
21:55And so
21:57that they wanted to still
21:58see the case. They wanted
21:59to talk to the pathologist.
22:01And so not all of
22:02them, but the neurosurgeon certainly
22:03did. So it was very
22:05simple.
22:06Here's an example at Children's
22:07Hospital where we have a
22:08pediatric neuropathologist on the bottom
22:10right with desktop screening software.
22:13We made
22:14monitors available in the OR.
22:16And that's doctor Wiley before
22:18he retired, our neuropathologist.
22:20So now the surgeons feel
22:22that they're still part of
22:23the action. They can see
22:24the case, talk to the
22:25person, just not in person.
22:30And because word got out
22:32and our neuropathologists
22:33were pretty good at doing
22:35frozen sections remotely,
22:37we got requests from outside
22:39of UPMC.
22:40And we've agreed to do
22:41that. We don't want to
22:42do too many because, as
22:43I've said, we've opened Pandora's
22:44box, and they're so busy
22:45now.
22:46And the hospital will not
22:47allow me to have too
22:49many neuropathologists.
22:50They already think I have
22:51too many neuropathologists.
22:52But now Merripath, for example,
22:54which is three hundred and
22:55seventy miles away in Indianapolis,
22:57we do their Frozen for
22:58them too.
22:59And why would I do
23:00that? Well,
23:02we don't make that much
23:03money through frozen sections. But
23:05what happens is now that
23:06we have a relationship with
23:07this group, we do their
23:08frozen.
23:09They send us many of
23:10their permanent cases. They send
23:11all their molecular testing to
23:13us, and we have a
23:13very good relationship. So we're
23:15their choice of lab to
23:16do all their work.
23:18And it's not easy, just
23:20so that you know, because
23:22now my pathologists
23:24need a license
23:26in a different state. They
23:27need to maintain that CME
23:28in that state. They need
23:30hospital privileges in a different
23:31hospital system. And so there
23:33are a lot of nuances.
23:35And, boy, you do not
23:36want to make a mistake
23:36when it comes to regulations.
23:38There was a lawsuit,
23:40just FYI, did not happen
23:42at UPMC, where a pathologist
23:44made a misdiagnosis
23:46called some something melanoma when
23:48it was a Spitz Nevis.
23:50And the reason the case,
23:52I mean, the pathologist lost
23:53the case was not because
23:54of the misdiagnosis,
23:56was the fact that they
23:57did not hold a medical
23:58license in the state where
24:00the the slide was scanned.
24:01Okay. So on that technicality,
24:05you cannot practice medicine if
24:07you don't have a license
24:08in that state.
24:09Now through the years, we've
24:10learned a lot of things,
24:11and I could stand here
24:12for another hour and give
24:14you lots of vignettes
24:16of how things were difficult
24:18and how things didn't work
24:19out that well. But I'm
24:20only going to share one
24:21vignette with you. Okay?
24:25And the scenario goes as
24:26follows. So why I'm telling
24:27you this is because you
24:29really need to use this.
24:31You're not gonna
24:32get this kind of experience
24:34through any book, through any
24:35vendor, or just having a
24:36quick conversation with someone at
24:38a peer institution.
24:39You really need to get
24:40the technology and use it,
24:42and you, you know, you
24:43will learn and people will
24:44become comfortable. And, unfortunately, you
24:46will learn from mistakes. So
24:47here's an example.
24:49We had an instructor
24:50at this particular hospital, and
24:52he was asked to do
24:53a frozen section on a
24:54Whipple for the margin.
24:56And there's the frozen section
24:57on the top. And he
24:59thought that's atypical. So he
25:00told
25:01the whipple surgeon,
25:04big deal whipple surgeon, okay,
25:06that it's atypical.
25:07And, of course,
25:08being a junior faculty,
25:10the surgeon said,
25:12I want a second opinion.
25:14And I don't just want
25:15anyone. I want the head
25:16of GI at UPMC.
25:18So, of course, we had
25:19to page the head of
25:20GI at a different hospital.
25:21We had to find her.
25:23We had to get her
25:24her to log on to
25:25her computer, and she didn't
25:25really know. Okay. Show me
25:27how do I log on
25:27and how do I look
25:28at it. And so she
25:29looked at the case and
25:30she said, I'm going to
25:31call that negative. Well, it
25:33came out permanent, which is
25:34there, positive for cancer.
25:37And, you know, that's not
25:39a good scenario.
25:40And it's not the technology's
25:42fault. But the point here
25:43is that you cannot just
25:45expect to ask people to
25:46use technology. And even if
25:48you validate it,
25:50you know, it's validated
25:51so that the CAP
25:53says it's safe. But you
25:55really need to make sure
25:56that people are competent, trained,
25:58and ongoing competency and monitor
26:00them. And so our neuropathologists,
26:02for example, are very good.
26:03They monitor
26:04concordance rates, deferral rates every
26:06month. They check with faculty.
26:08So if one of the
26:08faculty's deferral rate is too
26:10high, then
26:12doctor Julia Koffler, head of
26:13neuropathology, will have a conversation.
26:15Why are you deferring everything?
26:16Or why is, you know,
26:17your,
26:18frozen versus final, you know,
26:20out of whack compared to
26:21everyone else?
26:22So it needs to be
26:23part of your QA.
26:26Well, the next thing that
26:27happened was,
26:29well, you know, UPMC said,
26:30well, if we can share
26:31images, we should allow other
26:32people outside the system to
26:34send images to us for
26:35a second opinion.
26:37And, well, we decided to
26:39get into the consultation business.
26:42Wasn't my choice. Doctor Sam
26:44Musen, the executive vice chair,
26:45instructed me. Leron, you need
26:47to build this system. Because
26:49there wasn't anything we could
26:50buy back in the day.
26:51You know, now we have
26:51Path Presenter, Choristo, many tools,
26:53but there weren't many.
26:55So we built this IT
26:56system, And I'll tell you
26:57that we customized it for
26:59ourselves.
27:00And you guys know, Peter
27:01and John, customizing, you can
27:02get things to work perfectly
27:04for you, your your institution.
27:06And we had the world's
27:07best telepathology
27:08system, I believe, and still
27:10believe even though I've seen
27:11many others, that worked very
27:13well. And we received consultations,
27:15and the vast majority of
27:16our consultations came from China.
27:18There was a reference lab
27:19in China called KingMed,
27:21and they sent us ten
27:22thousand or more images a
27:23year.
27:24And they paid very well.
27:25We had a contract basis.
27:27And we paid the pathologists
27:29to do the work.
27:30But a couple of things
27:32about that. So one, yeah,
27:34you get really cool cases,
27:36rare cases, academically very challenging.
27:40But from an IT perspective,
27:41it's a pain in the
27:42butt.
27:43Okay? Because, you know, you
27:44need IT people to support
27:45this. Our IT people did
27:47not know how to speak,
27:48you know, Mandarin or Chinese,
27:49and all the servers were
27:50in Chinese.
27:52They had to work at
27:53different hours. So if we
27:53needed to connect something, make
27:55sure that their system was
27:56barcoded, I needed my IT
27:58analyst to wake up at
27:58four in the morning to
27:59meet with theirs. And you
28:00can imagine how well that
28:01went down.
28:03Culturally,
28:04very different the way we
28:05practice pathology and they practice
28:07pathology.
28:09Also, this reference lab, the
28:10main one, received
28:12samples from about nine hundred
28:14labs throughout China, small rural
28:16labs. And you can imagine
28:17how good their histology was.
28:19It was awful. And so
28:21people would call and complain
28:22to me, Leeron, the image
28:23is awful. Had nothing to
28:24do with the image. It
28:25was the slide site. Right?
28:26It was cut so thick
28:27and full of tissue folds
28:28and air bubbles and fingerprints.
28:30And I've so I never
28:31read many of those cases.
28:33Had nothing to do with
28:34the image.
28:35And so we we slowly
28:36had to get them to
28:37improve all the pre imaging
28:39factors.
28:41You know, it worked very
28:42well. But people ask me,
28:45and eventually
28:46politics killed that. So, So,
28:48for example, we cannot do
28:49business with China. They contact
28:51us and say we do
28:52not want to do business
28:52with the US for political
28:54reasons,
28:55and and we don't do
28:56that,
28:57that many cases. Now we
28:58can still get cases if
29:00people want to send them
29:00to me for teleconsultation
29:02work, But it's not a
29:03very
29:04viable
29:05business because it is very
29:07demanding from an IT perspective
29:09and even from a pathology
29:10perspective. So even if a
29:12pathologist is getting paid, say,
29:13forty five or fifty bucks,
29:15it takes hours to do
29:16some of these cases.
29:18They're very
29:19hard. You know, when a
29:19textbook says you get this
29:21you'll see this one in,
29:22you know, a hundred million.
29:23Well, we got one of
29:24those a week. Right? Because
29:26there were many such difficult
29:27cases that filtered our way,
29:29and it took hours and
29:30hours to do, and you
29:30had to consult
29:32a friend, and then you
29:33felt bad that I'm getting
29:34the forty five dollars and
29:35you're not, etcetera. So it's
29:37not really
29:38the best revenue generator.
29:40But then eventually what happened
29:41at UPC was I got
29:43contacted from people all around
29:44the world, from from Chile,
29:46Tibet.
29:47We know you guys do
29:48teleconsultation.
29:49We see it on your
29:50website, but we don't own
29:51a scanner because we can't
29:53afford it, but we still
29:54want to send you digital
29:55consults.
29:56And so how do we
29:57do that? And so at
29:58that time, if many of
29:59you remember, there were all
30:00these adapters that you could
30:01attach to your microscope, and
30:02you could put your smartphone
30:04on and take pictures.
30:06And so,
30:08we built this pocket pathologist
30:10so people could take pictures
30:12and upload it.
30:14The app is still available
30:15on the iTunes Store, but
30:17we do not support this
30:18because it became so demanding.
30:20And, again,
30:22if you wanna do this
30:24from a charitable point of
30:25view, makes a whole lot
30:26of sense. But if you
30:27wanna do this from a
30:28profitable
30:29profitable point of view and
30:30generate generate revenue, there's no
30:32way there's no way that
30:34a poor pathology lab in
30:36Tibet can afford US prices.
30:39And so when they, you
30:39know, found out how much
30:40we're going to charge for
30:41a consultation,
30:42like, they couldn't afford that.
30:44Alright?
30:45So many times we would
30:46just do it gratis.
30:49And it it also drove
30:51my IT people crazy, which
30:52is why I ended it.
30:53They got called all the
30:54time for help. How do
30:56I load this? How do
30:57I attach the camera? My
30:58adapter's falling. I can't focus.
31:00All of those kind of
31:00questions. It's just it was
31:02fun and gimmicky, but it
31:03doesn't really generate revenue.
31:05But I remember a really
31:06funny story is,
31:08one of our GI pathologists
31:09got twenty images sent to
31:11them.
31:13GI pathologist at Presbyterian Hospital.
31:15And it was because the
31:16person had a pancreatic neoplasm,
31:17and they weren't sure if
31:18it was adenocarcinoma
31:19or a neuroendocrine tumor or
31:21some combined tumor. It was
31:22just very confusing.
31:23And so they uploaded the
31:24images.
31:25But more than half the
31:26images
31:27were selfies because the pathologist
31:29had the camera facing the
31:30wrong way around.
31:32Right? So, you know, there
31:34was
31:35it's just it was just
31:36a huge headache. Nevertheless, we
31:37learned a lot from that.
31:38So what did we learn
31:39from our teleconsultation
31:43business? That there are not
31:45enough pathologists in the world,
31:46and there are certainly not
31:47enough expert subspecialized pathologists around
31:50the world. So if you
31:51wanna do this,
31:53you know, yes. People are
31:54desperate for help. But can
31:56they afford it? I don't
31:57know. You have to figure
31:58out a good deal with
31:59someone who will pay you,
32:00like, per capita
32:02or or contracted amount. But
32:03by case, I'm not so
32:04sure that makes a whole
32:05lot of, you know, money
32:06for you. But do you
32:08change patients' lives? Absolutely. What
32:10we found was in the
32:12beginning, the cases that were
32:13sent to us from were
32:14from pathologist to pathologist.
32:15So the pathologist was sending
32:17them to us. But then
32:17word got out because we
32:18made some major radical
32:20diagnosis that changed people's lives.
32:23I remember a case that
32:24came in as a sarcoma,
32:26and this patient was scheduled
32:27for an amputation.
32:28And and we did an
32:30AFB stain, and that's all
32:32it was. It was a
32:32mycobacterial spindle cell neoplasm.
32:34The patient just needed antibiotics
32:36for TB,
32:38not an amputation. And so
32:39eventually, clinicians
32:41realized that, wow, sending cases
32:42to UPMC in America,
32:44you know, is important when
32:46stakes are high. So then
32:47the clinicians asked for
32:49consults to get sent to
32:50us and eventually patients. Patients
32:53overseas were prepared to pay
32:54out of pocket to get
32:55a second opinion. And so,
32:57yes, it does work. But
32:58what the pathology
33:00clients valued most was not
33:01our diagnosis, but our education.
33:03The fact that we taught
33:03them about stuff. And our
33:04pathologists were generous in that
33:06they sent an article or
33:07they gave one or two
33:08references,
33:09about these novel entities and
33:10so forth.
33:11But required a huge commitment,
33:13big IT lift, lots of
33:15challenges, especially cultural. And I'll
33:17tell you because a lot
33:18of people ask me, we
33:19want to get into the
33:20consultation business.
33:23How do we get these
33:23digital slides? And it's much
33:25harder than that because here's
33:26a typical example. People want
33:28to send you a small
33:29round blue cell tumor. If
33:30they send you the digital
33:31slides, all you're gonna be
33:32able to tell them is
33:33it's a small round blue
33:34cell tumor.
33:35You're gonna need the tissue
33:36to run FISH or immunos,
33:39you know.
33:40Are you gonna look
33:42for, you know, subtle signs
33:44that this is a lymphoma
33:46and be sure, are you
33:47gonna commit a hundred percent
33:48on an HNE that this
33:49is Ewing's sarcoma and not
33:51do a follow-up EWS r
33:52one study?
33:54So
33:55from that point of view,
33:56they are not sending you
33:57tubular adenomas and basal cell
33:59carcinomas. They're sending you stuff
34:01that you know today is
34:02challenging or requires ancillary studies.
34:04How did we address that
34:05at UPMC?
34:06Well, what we told these
34:08labs sending to us, for
34:09example, in China is you
34:10need to up your game.
34:12Because if you're gonna send
34:13us certain,
34:15you know, specimens that we
34:16need we need ancillary studies
34:18on, and you cannot send
34:19us the tissue, for example,
34:20in China, they're not allowed
34:22to send the tissue, neither
34:23data over to us, then
34:25we need to make sure
34:25that your flow lab, fish
34:27lab, immuno lab has all
34:28of these stains.
34:30We even had their technologists
34:32come spend a month with
34:33us, train them, take our
34:35SOPs and procedures, go back
34:36and get them up and
34:37running, help them validate it.
34:39So when we wanted to
34:40run ancillary studies, we told
34:41them what to run, and
34:42they would run. We didn't
34:43get any of the revenue
34:45from it, but
34:46we maintained that teleconsultation
34:48business. Alright. That being said,
34:52I was a practicing side
34:53up pathologist, and I did
34:55not like the fact that
34:56I had to run around
34:57all over the hospital doing
34:58on-site evaluations,
34:59especially if I had to
35:00leave the building,
35:01and neither did my colleague
35:03Doctor. Sarah Monaco. We're like,
35:04what the heck? We're running
35:05to fifteen places in a
35:07day, and then we've got
35:07to get back, and then
35:08I've got a huge pile
35:09of slides. I've still got
35:10to sign out.
35:12And she said to me,
35:13you're the digital pathology guy.
35:14Figure it out. So
35:16at that point, no one
35:17was really doing telesitology.
35:19So we decided I decided
35:20then, you know what? Wherever
35:22we got cam wherever we
35:22got microscopes, I'm putting cameras
35:24on them, and we're streaming
35:25them. And I'm training all
35:26the
35:27cytologists, cytotechnologists
35:28how
35:29to stream images to us.
35:32And it makes total sense.
35:33So
35:34now we have
35:36twenty sites
35:37that I do telesitology.
35:39I still practice and do
35:41cytology service.
35:43I have four computer monitors,
35:44believe it or not.
35:46Told yours your your faculty
35:47you need more monitors, and
35:49I can do three on-site
35:50evaluations simultaneously at the same
35:52time.
35:53Yes. I multitask,
35:54super crazy, but I don't
35:55have to leave my office.
35:56I'm way more efficient.
35:58But the one thing that
35:59did happen was and and
36:00you need to know about
36:02this because as you go
36:03digital,
36:04your system is gonna tell
36:05you you need less and
36:06less pathologists.
36:08So what happened here is
36:10we got so busy that
36:11I requested,
36:13through hospital leadership that we
36:15need one or two more
36:16side of pathologists
36:17because we cannot cope with
36:18all the work.
36:21Then at the same time,
36:22I asked them to pay
36:23for the technology and put
36:24it in. And then when
36:25we got the technology and
36:26we were efficient, and they
36:27said, oh, you're publishing papers,
36:29and this is is we
36:30see how efficient you are
36:31and how you get all
36:31the work done. I was
36:33never allowed to hire the
36:34pathologists.
36:35Okay? Because the cameras
36:37replaced the pathologists,
36:38which is true. So I
36:39just want you to know,
36:41Angelique, that scanner in your
36:42office may be replacing a
36:44pathologist. I don't know. But
36:45at UPMC, the cameras did
36:46replace pathologists.
36:48Okay.
36:50So let's
36:52we then moved into whole
36:53site imaging at UPMC. Okay.
36:54And so currently, how do
36:56we handle whole site images?
36:57So I've shown you, yeah,
36:59sort of the way we
37:00deploy it. So let's start
37:01at the bottom.
37:02So we have scanners in
37:03our institution connected to our
37:05own Internet safe and secure
37:07behind our firewall, and those
37:08images are linked to the
37:09lab information system.
37:11That's a little,
37:13pile of circles there. So
37:15those are all linked. So
37:16those are live images in
37:17the LIS, and we can
37:19use those for primary diagnosis,
37:20tumor board, whatever we want.
37:22But we also scan a
37:23lot of slides,
37:25and we don't link all
37:26of those to the,
37:27to the LIS,
37:28and we use those for
37:29many other things:
37:31teaching, education, and so forth.
37:33We are able to accept
37:34wholesale images outside the system.
37:37I do that less and
37:38less now because of our
37:39experience through teleconsultation.
37:41It's not that profitable, and
37:42it's a big IT lift.
37:44But we are willing to
37:45do that, and I often
37:46do that if someone's asking
37:47as a favor. We have
37:48someone, they're desperate for a
37:50diagnosis. They know that we
37:51you have an expert at
37:52UPMC. Please, can they look
37:54at the case? Fine. We'll
37:55do that for you.
37:56We allow our pathologists to
37:58sign out remotely.
37:59So
38:00am I allowed to let
38:01people? USCAP is coming up.
38:03I know that many of
38:03our faculty will be there.
38:05Our system is validated for
38:06primary diagnosis. Can they sign
38:08out from their hotel room
38:09with the laptop?
38:10Yes.
38:12Can people work from home?
38:14Yes.
38:15And we do upload some
38:16of our images to the
38:17cloud. That's for AI analysis
38:19because some of the vendors
38:20that we contract with require
38:22the the AI to be
38:23run-in their environment, not our
38:26environment. And so, you know,
38:27that's
38:29not easy, but we do
38:30that. So that's kind of
38:31the topology of how how
38:33wholesale images do move around
38:34the
38:35system. Now
38:37about ten years ago, UPMC
38:39and GE went into business,
38:40and they started this digital
38:42pathology company called Omnix.
38:45And UPMC did this study,
38:47which was published,
38:48and they convinced themselves that
38:51if they install the Omnix
38:53digital pathology system,
38:55the whole sign imaging, and
38:56they roll it out over
38:57five years,
38:58and And they look at
38:59how will the system benefit
39:00from it, how much does
39:01it cost. And they actually
39:02went, and they looked at
39:03all the lawsuits UPMC had.
39:06They looked at one or
39:07two lawsuits.
39:08And who signed it out?
39:09Was it a generalist, or
39:10was was it a specialist?
39:12Like, did a dermatopathologist sign
39:14out the melanoma, or was
39:15it a general pathologist in
39:16a community hospital?
39:17They extrapolated that to the
39:18top ten cancers, and how
39:20much would that cost the
39:21system? And so they they
39:23that seventeen point that seventeen
39:25point seven million dollars is
39:26how much UPMC would save.
39:28That was their calculation. And
39:29based on that, they said,
39:31and I was there at
39:31the time doing informatics,
39:33Leron, your job is to
39:34implement this Omnic system and
39:36everyone has to do primary
39:37diagnosis.
39:39They didn't tell me that
39:41everyone who works at GPMC
39:42is gonna hate me and
39:43not wanna do it.
39:45Right? But that was what
39:46happened back in the day.
39:47The histology lab didn't want
39:49me bringing in these scanners,
39:50taking up space.
39:51The IT people were like,
39:53we know nothing about digital
39:54pathology. Like,
39:55why are we doing this?
39:57The pathologists
39:58did not like it either.
40:00Right? So that was it.
40:01I even had one pathologist
40:03who did his own time
40:04and motion study.
40:07He was a breast pathologist.
40:09He's a very famous breast
40:10pathologist today. But the breast
40:11pathologists, as you know, always
40:12say that they have the
40:13most slides per case. But
40:14that's because it's their own
40:15fault. They take so many
40:17sections per case.
40:21But he he proved that
40:23if he did his daily
40:25work with a microscope and
40:26glass slides, he was way
40:28more efficient
40:30than if he did it
40:31digitally.
40:32In fact, he was twenty
40:33five percent quicker.
40:35And he asked to have
40:35an appointment with me, him,
40:37and the chair of about
40:38Apartment at the time, and
40:39we met.
40:40And, I said, that's true,
40:42actually. Pathologists are very quick,
40:43especially very experienced, you know,
40:45getting that slide to move
40:46over the stage with a
40:47microscope.
40:47But then I asked him,
40:49did you take into account
40:51how long it takes the
40:53if we're counting turnaround time
40:54to get the case out,
40:55how long it takes from
40:56the histology lab
40:58to get you that slot
41:00if you add that in
41:01versus it was available immediately
41:03digits?
41:04And, of course, the minute
41:06he saw
41:07the turnaround time, not from
41:09his perspective,
41:10but from the lab's perspective,
41:11that the throughput,
41:13okay, that was obviously a
41:15deal breaker. So he realized
41:16that
41:17that doesn't make a lot
41:18of sense. And today, he
41:20keeps asking me, when are
41:21we getting more scanners? When
41:22are we getting more scans?
41:23Okay. That same pathologist.
41:26Okay. Lab also figured out
41:28that if we go slide
41:29less, basically,
41:30it would be way more
41:31lean and efficient because the
41:32spaghetti maps of, you know,
41:33making slides, taking them off
41:35to the immuno, bringing them
41:36back,
41:37etcetera,
41:38takes forever.
41:40You can imagine now it's
41:41even worse because we have
41:42less histotech. It's very hard
41:43to find histotechs,
41:45to staff all the labs,
41:46but we would be more
41:48efficient.
41:50And we did learn a
41:51bunch of things, and Doug
41:52Hartman, who worked with me
41:54at the time, published this.
41:54What we learned is, number
41:55one, all the pre imaging
41:57stuff is way more important.
41:58How you make the slide,
42:00how is it how quickly
42:01you can dry it before
42:02you can scan it, and
42:03so forth.
42:04That if you if you're
42:05asking pathologists to log into
42:07a digital pathology system, then
42:08a lab information system, then
42:10an AI system, they're gonna
42:11hate you. They just want
42:12one system. So we decided
42:14back into everything needs to
42:15go through the LIS.
42:16We figured out how to
42:17validate people, train people, and
42:20everyone asks us, are you
42:20going FDA? Are you using
42:22FDA? No.
42:24FDA does not regulate us.
42:26The FDA regulates the manufacturer.
42:28And we can choose to
42:29use an FDA
42:30approved or cleared system or
42:32not. In fact, of the
42:34one thousand I or one
42:36thousand plus antibodies we have
42:37in our immunohistochemistry
42:39lab, I I don't think
42:40any of them are FDA
42:41approved for everything we use
42:43them for. They're all off
42:44label.
42:46Plus,
42:47FDA system locks you down.
42:48You cannot change anything. And
42:50even if today I picked
42:51an FDA approved system, I
42:53would not be allowed to
42:53look at any of our
42:54archived images. If I wanna
42:56go look at someone's biopsy
42:57from a year ago, I
42:58would not be allowed to
42:59pull it into the current
43:00FDA system. So it doesn't
43:02really make sense
43:03for our use.
43:05Anyway, we installed these
43:08scanners throughout the lab, and
43:09we learned a lot about
43:10space,
43:11lean flow. The closer your
43:13scanner is to the slide,
43:15the better it is.
43:17We also figured about figured
43:19out about the workflow. Now
43:20many people think that these
43:22image management
43:24systems,
43:25drive the workflow. They're wrong.
43:27An image management system to
43:28me is like an access
43:30database or an Excel spreadsheet.
43:31It manages the images very
43:32well if you're an administrator.
43:34But if you're a pathologist,
43:35no.
43:36You we do not
43:38manage our
43:39cases by slides and blocks.
43:41We manage them by case.
43:43So you really need a
43:44case management system.
43:45Parts, how many, you know,
43:47how many parts are there?
43:47How many slides per part?
43:49How many do need immunos?
43:50How many of those need
43:51an immunosistochemistry
43:52stain. When the stain comes
43:54back, you don't want it
43:55right at the bottom. You
43:56want it back in that
43:57part where it belongs. And
43:59so we worked very hard
44:00with our IT people to
44:02get the workflow to work
44:03out well.
44:05We were also a Copath
44:06user. We're now half an
44:08Epic user. By May ninth,
44:10we'll all be Epic users.
44:11But you can see that
44:13running it through the LIS
44:15makes a whole lot of
44:15sense. And the more we
44:16did this, the more pathologists
44:17accepted it because this was
44:19their workflow. You could look
44:20at this particular case as
44:21part one, part two. I
44:23can see of that part,
44:25three other slides were scanned,
44:26three were not. You can
44:27view one side. You can
44:28view them all. But it's
44:29all through the LIS. You
44:30order a recut. The recut
44:32will get scanned. Come back.
44:33You'll see, oh, it's exactly
44:34in that part. You're in
44:35immuno. You don't have to
44:36go anywhere. You can get
44:37it right done. You need
44:38to now do the report.
44:39You're all in the same
44:40system. So we preferred that
44:41at UPMC.
44:43What we also learned there
44:45is that
44:47currently the way Histo Labs
44:49work is stuff gets assembled,
44:51slides are made, paperworks match,
44:53they get onto flats or
44:54trays, they get delivered to
44:55mailboxes.
44:57And, you know, walk by
44:58and you still have mailboxes
44:59with slides in them. Right?
45:02People felt that going digital,
45:04that all goes away. It's
45:05not true. And we learned
45:07that.
45:08There's still
45:09this kind of work that
45:10needs to be done, administrative
45:11work, but it's in the
45:12digital system. Someone still has
45:14to go in and say,
45:14doctor Desik's on service today.
45:16Next week, she'll be out.
45:17This has to go to
45:18someone else. Someone's gotta go
45:19in and do all that
45:20work and make sure that
45:22the QC gets done.
45:23Now the one thing I
45:25don't agree with the way
45:26some labs that have gone
45:27fully digital are doing their
45:28QC is, how do we
45:30know every image is okay
45:31and every image is in
45:32focus? They're saying, oh, we'll
45:33look at ten images for
45:35each batch. But that sort
45:36of is counter
45:38to automating the whole system.
45:40Now there's AI
45:41that does your QC checks,
45:42and we're testing a few
45:43of those different systems. I
45:45don't really want people to
45:46check that. But what did
45:48change is
45:49the follow just got very
45:50used to
45:51knowing when they had work,
45:52when their mailbox got full,
45:53or someone delivered them stuff,
45:55which is a push system.
45:57What happened when we were
45:58using the omnich system, they
46:00didn't know that their cases
46:01were building up, which is
46:02a pull system. You gotta
46:04go in and you gotta
46:04go log in and see,
46:05see, oh my goodness. I
46:06got thirty five cases, and
46:07it's already eleven o'clock. I
46:08better get started. So that
46:10didn't happen.
46:11And I did learn this.
46:13So I'd asked money for
46:15money to buy the IT
46:16system, buy the hardware, buy
46:17the software.
46:19And they gave me all
46:20these buses.
46:21All these buses arrived and
46:22got lined up, and then
46:24I had no bus drivers.
46:26Okay. Who's gonna scan the
46:27slots?
46:28I'd assume that the hospital
46:29would say, okay. You can
46:30use the histotech time and
46:32or, an admin assistant that
46:34time will train them. We'll
46:35hire you a bunch of
46:36scan tags. No.
46:38If it wasn't budgeted for,
46:41the those were just gonna
46:42sit there and, you know,
46:44and collect dust, which happened
46:46for a while. So I
46:47told them, look. We spent,
46:48you know, seven million dollars.
46:50We do need to, use
46:51the technology.
46:52Okay?
46:53Now I know, having done
46:55that, is you need people.
46:56You need IT analysts. You
46:57need a business analyst. You
46:59need a director of digital
47:00pathology. You need Scantecs.
47:02Otherwise, you're just gonna have
47:03a fleet of buses that
47:04are gonna go nowhere. Okay?
47:08And how do you sign
47:09out? So when you speak
47:10to all the digital pathology
47:11companies, they'll tell you, yeah.
47:13Sign out. It's easy. Look
47:14how we do it in
47:14the electronic system. But that
47:15is not how we sign
47:16out for real. You know
47:18as pathologists.
47:20Sign out,
47:21many times this means we're
47:23looking at the case at
47:24a microscope, either on my
47:26own microscope
47:27or a multi headed microscope
47:28in a conference room, and
47:29then I'm gonna go back
47:30to my office and then
47:32sign out. And, like, digital
47:33pathology vendors are confused. Like,
47:34what are you talking about?
47:36Sign out is when you
47:37sign out the case. No.
47:38Sign out is when you
47:38look at the case, you
47:40figure out how you're gonna
47:40report it, and then you
47:42may electronically sign it out.
47:43So you have to ergonomically
47:45figure out how pathologist offices
47:47are gonna look, where are
47:48these digital pathology monitors gonna
47:50go.
47:51When you're signing out in
47:52a conference room, where's everyone
47:53gonna be, if you wanna
47:55put TVs on the wall.
47:56Well, the aspect, the way
47:57that the image appears is
47:58very different to what it
47:59looks like on the monitor.
48:00And if you you chose
48:01an FDA sys approved system,
48:04putting it up on the
48:05monitor is certainly not part
48:06of the pixel pathway that
48:07the FDA approved, and it
48:08becomes quite complicated. So that's
48:10something you really have to
48:11figure out.
48:12And you've got a lot
48:13of conference rooms around you
48:14and and, you know, ergonomically,
48:16you have to figure that
48:17part out.
48:18But, anyway, we went. We
48:19were scanning away, doing primary
48:21diagnosis with the OmniC system.
48:22And then
48:24the CIO called me and
48:25said, Leron, you're filling up
48:27our servers.
48:29Like, they hadn't planned for
48:31us. He said, I just
48:32look at this.
48:34MRIs, CT scans, PET scans,
48:36they take up hardly any
48:37space, but you guys are
48:39filling up the server at
48:40this enormous rate. Okay? And,
48:42you know, back in the
48:42day, we didn't really think
48:43much about data storage and
48:45how big a whole site
48:45image is, but that's a
48:47big deal. And for most
48:48labs, actually now when they
48:49are budgeting and IT get
48:51wind of how much you
48:52have to pay for data
48:53storage and you want to
48:54back it up as data
48:56recovery,
48:57for disaster recovery,
48:59you have to factor in
49:00that storage is very expensive,
49:01especially if you want to
49:02keep the images.
49:04If you wanna just put
49:05move them to cold storage
49:06on a disk, well, then
49:07it's like a glass slide.
49:08When you request an image,
49:09it'll take days, which makes
49:10no sense.
49:12Now I wish someone had
49:13told me this.
49:15After all the years we
49:16were scanning images, we got
49:18the Omnic system, we got
49:19the Aperio system, we had
49:20a Zeiss system, we had
49:21an Olympus system, a Hamamatsu
49:22system,
49:23and each
49:24vendor had their own proprietary
49:26file format.
49:27And we still had them
49:28all saved. Just the other
49:30day, someone had asked me
49:31for seventy thousand GI polyps
49:33from the Omnic system.
49:35And I said, you can
49:36have them, but they said,
49:37how am I supposed to
49:37view them? I don't have
49:38the viewing software or you
49:40know? So they're not saved
49:41in any standard way. And
49:42so the question is,
49:44should we just do them
49:45by DICOM now going forward?
49:46Because DICOM is a standard.
49:48Radiology adopted that. Should we?
49:49And, yeah, I would say
49:51we could do that.
49:52The problem
49:53is, will every AI vendor
49:54we want be able to
49:55use DICOM? Wanna do image
49:57analysis for ERP or HER2,
49:58they're gonna tell, no. We
49:59don't do it on DICOM
50:00images, etcetera. So the field
50:02has to move forward.
50:04Nevertheless, UPMT has advanced. We
50:07replaced the Omnic system with
50:08the Leica Aperio system.
50:11It's validated for primary diagnosis.
50:13And now we have people
50:14signing up from home. Not
50:15too many.
50:16Here's a perinatal pathologist, doctor
50:18Bob Benden. He lives in
50:19Florida. He's just recently retired.
50:22I brought him out of
50:23retirement to sign out for
50:24placenta cases because we have
50:25so many placentas,
50:27in Pittsburgh. And what actually
50:29happened was at the time,
50:30a lot of politics in
50:31the states around us where
50:32women could not have abortions,
50:33etcetera, they all came to
50:35Pennsylvania,
50:36and I could not cope
50:37with the number of placentas
50:38and POCs that had to
50:39be signed out. And so
50:40Bob said he would. He
50:41wouldn't come back and live
50:42in Pittsburgh. It's too cold.
50:43His wife hated the cold.
50:45So he said he will
50:46sign out. Plus, you can
50:47see there's the the sea
50:49behind him. So in the
50:50morning, he'd wake up, have
50:51breakfast, go for a walk
50:52on the beach, then he'd
50:53come sign out the centers
50:54all day. And our residents
50:56loved him because he took
50:57time and explained it to
50:58them, and they would join
50:59him virtually and they could
51:00send it. So
51:02he told me I have
51:03to pay for his license
51:03in Florida
51:05and Internet access and all
51:06of that. But what we
51:07do have is we have
51:08a team now who can
51:09deploy a workstation. So if
51:10someone wants to work from
51:11home, we take care of
51:12it. We send them a
51:13workstation, get it set up,
51:14barcode monitor. You have to
51:15use our computer.
51:17You have to VPN in
51:18and and you have to
51:19follow all our rules.
51:20And so we have some
51:21pathologists signing out. I even
51:22have a pathologist,
51:24Marta Miniverni, in Italy. She
51:25signed up from home in
51:27Italy,
51:28for extended periods of time.
51:31But
51:32what we did learn
51:34is
51:35I again opened another Pandora's
51:37box because a lot of
51:38senior pathologists asked me, hey,
51:39Lauren. I was gonna retire.
51:41Can I go part time
51:42and work from home? And
51:44I didn't want a whole
51:46department working from home because
51:47there are downsides to that.
51:49But also, if people do
51:51work from home,
51:52they do help us cover
51:53the service, cost coverage, etcetera.
51:55And so I found that
51:56the most
51:58popular phenotype of pathologists who
52:00asked me to sign up
52:01from home are
52:03either those who are
52:04one to five years in
52:05practice
52:06because
52:07hard to find care, especially
52:09if two people are working
52:10and then there's no one
52:11taking care of the kids.
52:13I was joking with doctor.
52:14So, Nad, if
52:16If you're signing out from
52:16home and taking care of
52:17the kids, like, are you
52:18sure the kids aren't all
52:19over your computer workstation?
52:23I've never asked that question.
52:26So people who want to
52:27retire and people who have
52:29childcare needs
52:30or other needs work from
52:31home. And I have so
52:32far, I have about six
52:34five or six people.
52:36The the sixth one is
52:37because when she's in Italy,
52:38she signed up, but when
52:39she's back, she comes into
52:40work.
52:42But I don't want the
52:43whole department to work from
52:44home. As you know, that's
52:45not the best thing.
52:47So we learned a whole
52:48lot of,
52:50stuff about
52:51going to primary diagnosis.
52:53If you are not married
52:54to your vendor and you
52:55don't like your vendor, you're
52:56in big trouble. This is
52:57a bad marriage because there's
52:59a lot of things that
53:00have to happen. If your
53:01pathologists are not willing to
53:03do this, it's gonna be
53:04an uphill battle.
53:06I would suggest you don't
53:07do it as a big
53:07bang and do it right
53:08away. What we have found
53:10is if you pick
53:12areas where small pieces of
53:14tissue like pediatric pathology,
53:17autopsy service, and then work
53:18your way up.
53:19I'll tell you a funny
53:20story that a colleague of
53:21mine decided, Dara, I'm not
53:22doing it your way. I'm
53:22not doing it incrementally. It's
53:24just too painful, especially if
53:25you have a hybrid system,
53:26half glass,
53:27half digital, just so difficult
53:29for the lab. We're going
53:30fully digital. And then one
53:31of his pathologists said, I'm
53:33quitting because I'm not doing
53:34it. And the reason was
53:35because
53:37he couldn't see the images.
53:38He didn't know he needed
53:40glasses because he could
53:41he could focus with his
53:43microscope, sadly, but not on
53:44a computer monitor.
53:46And so once they figured
53:47that out, he didn't quit
53:48and, you know, it was
53:49okay.
53:51Alright.
53:52Now not every slide in
53:53the world can be scanned.
53:56These are slides people asked
53:57me to scan, old AFIP
53:59slide, broken slide, double cover
54:01slip, etcetera. You know, the
54:02scanner will not scan everything
54:03around, so you will still
54:03need a microscope around somewhere.
54:03And this garbage in, garbage
54:03out principle is
54:09important, as I told you.
54:09Everyone
54:11needs to know. Your resident
54:12set of grossing, your surge
54:13techs, peers need to know
54:14that whatever I'm cutting this
54:16piece of tissue, putting in
54:17a in a cassette is
54:19ultimately gonna be scanned and
54:20be on someone's computer monitor.
54:22So everyone needs to be
54:23mindful. So, for example,
54:25if you have all of
54:26these pieces of tissue,
54:28completely fills the block, when
54:30they open the cassette and
54:31it's about to explode, that's
54:32gonna go all over the
54:33slide. You don't want
54:35that. When your histo technologists
54:36are arranging stuff, for example,
54:38you don't want them to
54:38put tissue all over the
54:39place. You want it really
54:41well arranged, close together because
54:43if it's close together, it's
54:44easy to scan. You're not
54:45scanning all this white space,
54:47storing it. When someone's looking
54:49at it on a monitor,
54:50you don't want, you know,
54:51them to get carpal tunnel
54:52at the end of the
54:52month. You know, you want
54:54it to be easy.
54:56I finally convinced our geopathologists
54:57that they need ribbons and
54:59ribbons and ribbons of everything
55:00on, you know, slide after
55:02slide after slide. Even our
55:03geopathologists.
55:04Less slides, less ribbons,
55:06and they agree it actually
55:08saves a lot of histotech
55:09time.
55:11Even
55:12our histo lab now have
55:14pristine slides, pristine barcodes,
55:16no fingerprints,
55:17no mounting media. That all
55:19impacts the quality of the
55:20image.
55:21And I proved that to
55:22myself. If if if you
55:23cannot focus, can you imagine
55:25what an image algorithm is
55:26gonna look, try and deal
55:27with it? So this was,
55:30an image analysis algorithm to,
55:32you know, quantify HER2. You
55:33could see the green is
55:34HER2 one, then there's HER2,
55:36and HER3 is the red.
55:38And the more multimedia and
55:39blurry it is, I can
55:41make any HER2
55:42zero.
55:43If it's blurry, the algorithm
55:44will make it zero for
55:45you. And how would you
55:46know?
55:48But do not be OCD
55:49about it. Right? That's it.
55:51Minor artifacts don't matter, which
55:53I've told our pathologists.
55:55Hey. Yeah.
55:56A fingerprint there, it's not
55:57the end of the world.
55:58You can see that you're
55:59dealing with a leiomyosarcoma
56:01because there's a hair under
56:02the coverslip. You know, you
56:03cannot refuse to sign out
56:04the case. Right?
56:07But there are some things
56:08that are a big deal.
56:08This is a core needle
56:09biopsy of a of a
56:11breast biopsy.
56:12This has got invasive dachal
56:14carcinoma. It went for HER2
56:15staining. And you don't see
56:17the HER2. It's a HER2
56:18zero case, basically.
56:20But the
56:21the image algorithm came back
56:23as three plus because it
56:24counted those bubbles under the
56:26coverslip
56:27slip as membranous staining. And
56:27if a human never checked,
56:29well, then it would be
56:30a mistake.
56:32So the next thing we
56:33did is open up an
56:34image analysis lab where we
56:35do quantitative image analysis. It's
56:37as simple as as Facebook
56:39does,
56:40classify the things in the
56:41image, detect them, segment them
56:42out, count the features.
56:44How much staining, what's intensity,
56:46etcetera. And so we do
56:48that now. But we have
56:50changed, and our image analysis
56:51lab runs like a CP
56:53medicine lab. We do Levy
56:55Jennings parts. We check for
56:56the CV. Is there drift
56:58in the algorithm? Is there
56:59drift in the immunostain?
57:01Yeah. I took two algorithms
57:02that were running and just
57:04ran them on the same
57:04twenty cases, and they were
57:06completely different for the ER
57:07and KR sixty seven. People
57:09would assume that, oh, the
57:10computer is always right. Well,
57:11it's not always right unless
57:13you monitor it.
57:15We've now moved into computational
57:17pathology. There's a fancy definition
57:19for computational pathology, but it's
57:21really AI mostly applied to
57:22pathology.
57:23And you can see in
57:24PubMed,
57:25there are thousands and thousands
57:27of articles
57:28about computational pathology. So if
57:30you're asking when should we
57:31do AI,
57:32that train has left that
57:34station long ago. You should
57:35have already been doing AI.
57:37Okay.
57:38So we asked
57:40AI, what would a pathologist's
57:41life be like using AI?
57:43And this is what it
57:43said. That was the only
57:45prompt.
57:48Looks like some of the
57:48offices in our hospital still
57:50are like that. I'm not
57:51gonna blame you.
57:53But in a digitally well
57:55organized world, you know, obviously,
57:56we could be much better.
57:58And you can use AI
57:58for anything you want.
58:00Go detect rare things like
58:01bugs,
58:03count mitoses,
58:04make diagnosis of cancer, make
58:05predictions, or use it for
58:07research.
58:08I got tired of screening
58:09AFB stains myself,
58:11and I never trusted the
58:12residents. The fellow said, Yeah,
58:13there's nothing there. Then I
58:14would look with them, and
58:14in one minute I'll find
58:15a mycobacterium.
58:17So we built our own
58:18AFIB screening
58:20algorithm. It's been running now
58:21for many years,
58:22and it detects the AFB.
58:24It's much more accurate and
58:25quicker.
58:27Now there are three types
58:29of pathologists.
58:30There are those who use
58:32this only. They don't want
58:33the slide anymore.
58:34There are those that wait
58:35for the slide. They trust
58:37the issue. They want to
58:37check themselves. And there are
58:39those that do not use
58:40this at all. Only wait
58:41for the slide. And so
58:43when I asked some of
58:44those pathologists, why do you
58:45not trust and use the
58:47AI? And you know what
58:48the main reason was?
58:49This.
58:53It's not integrated.
58:55If you're gonna use something,
58:56you want it nice and
58:57integrated and easy to use.
58:58They do not want it.
58:59They got an email notification.
59:01AFV is ready for your
59:02case, this case number. Now
59:03they gotta go log in
59:04to Copath,
59:05get into the case. It
59:06said, if it was all
59:07integrated,
59:08they would use it more
59:09often.
59:12When we we installed our
59:13first AI algorithm about six
59:14years ago, it was IBEX
59:16for prostate cancer.
59:18I asked the CAP, and
59:19I was on the CAP
59:20committee and many others, how
59:21do how do we validate
59:22this? We knew about whole
59:23site imaging, but how do
59:24we validate
59:25AI? How many cases? What
59:27are we checking for? What
59:27are parameters?
59:28They did not know. No
59:30one actually
59:31could give me advice.
59:33But what I did know
59:34is that I'm not gonna
59:35do analytical validation like a
59:36vendor does. You know? Build
59:38the algorithm, hold out a
59:40whole bunch of stuff. That's
59:41like building a car. I
59:42don't want that.
59:43I know that I cannot
59:44use technical verification.
59:46That's like going to a
59:47car dealership and the the
59:49car salesman showing you all
59:50the features and, yeah, we'll
59:51calibrate it and tweak it.
59:53No. That's no good. I
59:54need to test drive it
59:56so that I know it
59:57works.
59:58And how many cases? I
59:59don't know. So what we
01:00:00did, just so that you
01:00:01know so for prostate, I
01:00:03took three thousand slides,
01:00:06turned about so hundred cases
01:00:07with three thousand slides,
01:00:10scanned them,
01:00:11had them all analyzed by
01:00:13AI,
01:00:14and then we had,
01:00:15compared it to the pathologist
01:00:16diagnosis. And there were three
01:00:18zero discrepancies.
01:00:19Cancer and no cancer, grading
01:00:21differences, perineural invasion or not.
01:00:23And the pathologist at the
01:00:24time told me that, you
01:00:25see, AI is not right
01:00:26all the time. But the
01:00:27truth is all three of
01:00:28those cases were correct. The
01:00:30AI was correct. We adjudicated
01:00:31that by, you know, consensus,
01:00:33another expert, and immunohistochemistry.
01:00:36And here's an example where
01:00:37the pathologist originally felt this
01:00:39was negative. AI thought it
01:00:40was three plus three, and
01:00:41the AI was correct based
01:00:43on immunohistochemistry
01:00:44staining. And the xAI means
01:00:46explainable AI. Our pathologists really
01:00:47want to see what's going
01:00:48on with the AI. So
01:00:49by overlaying a heat map
01:00:51makes the AI explainable. There's
01:00:53the cancer. The red means
01:00:54it's we think it's malignant,
01:00:56etcetera.
01:00:58How do you install AI?
01:00:59That's a different issue. Do
01:01:01you do it upfront? Do
01:01:02you do it later? Initially,
01:01:04for example, the pathologist
01:01:05said, do it once we've
01:01:06done the case just as
01:01:08a quality check, but but
01:01:09they soon realized that that
01:01:10makes no sense. So much
01:01:12more efficient upfront. So now,
01:01:13for example, our GU group,
01:01:14by the time they get
01:01:15there in the morning, everything
01:01:16scanned, run by AI, preliminary
01:01:19report generated.
01:01:20Even the fellow is way
01:01:21more efficient, so their service
01:01:22is much more efficient.
01:01:24But how we use AI
01:01:25is different, and so because
01:01:27the hospital did ask me
01:01:29this. At the academic hospitals,
01:01:31my academic pathologists are a
01:01:33little insulted if I tell
01:01:34them that it's to check
01:01:35check accuracy because many of
01:01:36them tell me, Lauren, you
01:01:37know, people send me consults,
01:01:38etcetera. You know, I don't
01:01:40need that checked. What they
01:01:41really want is the efficiency.
01:01:43Save time so they can
01:01:44have time to write papers,
01:01:45do admin
01:01:46academic work. Whereas in the
01:01:48community, and we have many
01:01:50community hospitals, they do wanna
01:01:51know that they didn't over
01:01:53or under call anything. So
01:01:54it's the exact same algorithm.
01:01:55They're much more happier having
01:01:57AI check everything. And, actually,
01:01:59I'm happy that these general
01:02:00pathologists are not overwhelming our
01:02:02academic pathologists with all these
01:02:04cases. You know, prostate with
01:02:05ASAP,
01:02:06you can handle it on
01:02:07your own. You do not
01:02:08have to send it to
01:02:09a subspecialist.
01:02:10But a lot of questions
01:02:11have come up over time.
01:02:12You're on which apps are
01:02:13you using? How are you
01:02:15deciding only if it's FDA
01:02:16or if other people are
01:02:17using it? How are you
01:02:18paying for it? Should the
01:02:19residents and fellows use it
01:02:20or not, etcetera?
01:02:22And you will learn all
01:02:23of those over time.
01:02:25But I do feel that
01:02:27we should be doing more
01:02:28with AI. You know, not
01:02:29just having AI do exactly
01:02:30what pathologists do, same grading,
01:02:32WHO,
01:02:33Bethesda. We should really be
01:02:35looking into a crystal ball
01:02:37and making advances.
01:02:38And so that's where we
01:02:39should go. But
01:02:40I decided I cannot do
01:02:42this on my own, so
01:02:43I created a computational pathology
01:02:44and informatics division. Doctor Hooman
01:02:46Rasheedi, who trained here at
01:02:47Yale now, he runs that.
01:02:50Matthew Uhanna is the vice
01:02:51chair of informatics, helps with
01:02:53all of this process.
01:02:55And Ibrahim,
01:02:56a recurant, runs the image
01:02:57analysis lab.
01:02:58And they're all competent.
01:03:01I open up an AI
01:03:02center now called our computational
01:03:03pathology and AI center.
01:03:05They build AI. They validate
01:03:07AI. And I I they
01:03:08don't like it, but I
01:03:09tell them they're my oil
01:03:10refinery.
01:03:11They take all the crew
01:03:12data from lab medicine,
01:03:15seat,
01:03:15and as well as images,
01:03:17and then make whatever jet
01:03:18fuel I need. I need
01:03:19them to do this. I
01:03:20need them to do that,
01:03:21and they do that,
01:03:22along with an academic mission.
01:03:25The one thing I found
01:03:26over time is it's very
01:03:27hard to work with vendors.
01:03:28I used to vendors would
01:03:29say, no. We have a
01:03:30new scanner. We have a
01:03:31new system. Can you test
01:03:32it for us? But I
01:03:33had all these rules in
01:03:34our hospital. Ninety days,
01:03:37supply chain management won't allow
01:03:38this, and so I decided
01:03:40I'm not going through that
01:03:41process anymore. So I opened
01:03:42up a digital pathology research
01:03:43center.
01:03:44I called it research. So
01:03:45it's under research. Hospital have
01:03:47no sort of,
01:03:49you
01:03:50know,
01:03:51say in how long anything
01:03:52stays or which vendor comes.
01:03:54You know, we make sure
01:03:55we have ethical clearance. We
01:03:56have compliance, IRBs.
01:03:58I had so many
01:04:00companies come to me, digital
01:04:02pathology, AI,
01:04:03and even novel imaging companies,
01:04:05Muse and others.
01:04:06I had to open up
01:04:07an office of collaborative pathology.
01:04:09I have a medical director
01:04:10who runs that. I have
01:04:11a lawyer now, an administrator.
01:04:12I have a compliance officer.
01:04:14I have a PhD scientist
01:04:16who checks the scope of
01:04:17work. And so we've already
01:04:18done forty projects,
01:04:20multimillion dollar
01:04:22operation, and that money all
01:04:23comes into the department.
01:04:25Okay? And And everyone loves
01:04:26it because they get access
01:04:27to technology. If we like
01:04:29it, we'll translate it and
01:04:31move it over into the
01:04:32clinical operation.
01:04:34And lastly, what we've realized
01:04:35at UPMC is that I've
01:04:37had algorithms that we've built
01:04:38ourselves, but they only work
01:04:39at UPMC. Like that acid
01:04:41fast bacillus algorithm. I've had
01:04:43a pathologist send me and
01:04:44ask me if we would
01:04:45run it, run her images
01:04:46for us.
01:04:48She wanted to quantify them
01:04:49for her research project. And
01:04:50we ran them, and we
01:04:51couldn't find the AFB.
01:04:53So just because we build
01:04:55it doesn't mean it's generalizable.
01:04:56So we have decided that
01:04:58we are building a consortium
01:05:00which
01:05:01is underway.
01:05:02We are working with partners
01:05:03around the U. S. Globally,
01:05:05people from Chile, Mexico,
01:05:08Japan,
01:05:10Tata Memorial Cancer Center, etcetera,
01:05:12who are all providing us
01:05:13with data so that we
01:05:14can build much better AI
01:05:15algorithms that can generalize for
01:05:17everyone and also meet all
01:05:19their priorities.
01:05:20So at the end of
01:05:20the day, that's the message
01:05:22I tell everyone I work
01:05:22with.
01:05:24I'm not replacing them
01:05:25yet,
01:05:27but I'm just giving them
01:05:28the tools to make their
01:05:29job easier. Right? Whether it's
01:05:31screening Pap tests or whether
01:05:32it's looking at corneal lobsters
01:05:34of breasts, prostate, etcetera. So
01:05:36I'll leave you with this
01:05:36final quote from Darwin.
01:05:38It's not the strongest of
01:05:39the species that survives nor
01:05:41the most intelligent,
01:05:42but it's the one that
01:05:43is most adaptable to change.
01:05:45So you have to be,
01:05:46you know, willing to change.
01:05:47So sorry if I went
01:05:48over for a little bit,
01:05:49but thank you for your
01:05:50attention.
01:05:55Doctor Daisig, is there time
01:05:56for questions or no? No?
01:05:58We have time for questions.
01:06:00Oh, okay.
01:06:03Any questions?
01:06:05First
01:06:09question.
01:06:21So when kids go and
01:06:23get their car license today,
01:06:24do we make them go
01:06:25with shift stick or automatic?
01:06:27When kids get their car
01:06:29licenses today, are we gonna
01:06:30make them all go with
01:06:31a shift stick so that
01:06:32just in case they have
01:06:33today, you're gonna use a
01:06:34shift stick, or is it
01:06:35okay to get your car
01:06:36license with an automatic vehicle?
01:06:38For me. The entire transmission
01:06:40of the solar system is
01:06:41still keeping That's right. So
01:06:43and I did this experiment.
01:06:44So our GU fellow was
01:06:45not allowed to our GU
01:06:47fellow was not allowed to
01:06:47use AI
01:06:49each year. But when I
01:06:50got there as the chair,
01:06:51I told him it makes
01:06:52no sense.
01:06:53Actually, what we're doing is,
01:06:55because they were saying we
01:06:55would deskill the the GEO
01:06:57fellow like you're saying. They
01:06:58would not be able to
01:06:59be as good as previous
01:07:00fellows to find the cancer
01:07:01and actually grade it themselves.
01:07:04And I told them, actually,
01:07:04what we're doing is, we
01:07:06are deskilling them to go
01:07:07practice in the future because,
01:07:09they would who's gonna train
01:07:11them to do AI, especially
01:07:12if we have AI around?
01:07:13We need to upscale them.
01:07:15And so I had a
01:07:16fellow who had done six
01:07:17months of his fellowship, GU
01:07:18fellow. And then I asked
01:07:19him, would he be willing
01:07:20to switch halfway through? And
01:07:22we'd actually document everything he'd
01:07:24done. So we documented everything
01:07:25he'd done for six months,
01:07:27his diagnoses compared to, you
01:07:28know, it's, the ones that
01:07:30sign out, how efficient he
01:07:31was, how efficient the service
01:07:33ran, etcetera. And And then
01:07:34we switched, and we documented
01:07:35everything. And so what had
01:07:37happened was, firstly, the service
01:07:38was way more efficient. Okay?
01:07:40Everyone could sign out by
01:07:41eight eight or nine AM
01:07:43in the morning. Everything was
01:07:44done, screened. All the cases
01:07:46that needed immunos were already
01:07:47sent off. They came back
01:07:48the same day because they
01:07:49met the cutoff.
01:07:50Okay.
01:07:53The fellow
01:07:54got much better. Okay. Because
01:07:57the fellow looked at the
01:07:57case and then turned on,
01:07:58you know, the heat heat
01:07:59maps afterwards. He didn't wanna
01:08:00ruin his ability not to
01:08:02be able to try and
01:08:02make diagnoses.
01:08:05And then guess what happened
01:08:06every time there was an
01:08:07issue, technical
01:08:09AI, whatever?
01:08:10Who do you think the
01:08:11champion was that they went
01:08:12to? All the attendings. The
01:08:13fellow.
01:08:15The fellow was so good
01:08:16that they asked me to
01:08:17hire him. He's now faculty.
01:08:18He's been on faculty for
01:08:19two years. Okay? And he's
01:08:20still the AI champion.
01:08:21And now he does GU
01:08:23research.
01:08:24And guess what he uses
01:08:25to do his GU research?
01:08:27He doesn't lean over a
01:08:28microscope doing it. He uses
01:08:29the AI to do all
01:08:31different things, spatial
01:08:32biology and different things.
01:08:34It's a completely new microscope
01:08:36for him. And so
01:08:38I would say, I think
01:08:39it would be detrimental if
01:08:41you have the tools and
01:08:41technology not to let the
01:08:43people use it. I'm not
01:08:44gonna force people to use
01:08:45a shift stick because that's
01:08:46just how everyone used to
01:08:47do it. So that's my
01:08:49opinion.
01:08:50Okay.
01:09:30So the answer is no
01:09:31to both your questions. I
01:09:32have not tracked people's,
01:09:34RBUs. I tracked them for,
01:09:37quality purposes or when to
01:09:39hire, but not for personal
01:09:40productivity. But now that you've
01:09:41told me, yeah, I think
01:09:42I'm gonna do that.
01:09:44I think it's hard because
01:09:45probably that will be Yeah.
01:09:47At some level. Yeah. So
01:09:49I think I'm gonna do
01:09:49that. The groups that use
01:09:50AI, and a good time
01:09:52so, you know, we're we're
01:09:53now doing all the PAP
01:09:55tests with the Genius AI
01:09:56system. So, it will be
01:09:58a good thing to monitor
01:09:59actually that we have,
01:10:02twelve side of pathologists that
01:10:03do PAPs. That's how I
01:10:04wanna monitor actually the RVUs
01:10:06before, RVUs now, see if
01:10:07they're quicker. The problem is
01:10:09I I'm reluctant to share
01:10:10that data because, you know,
01:10:11the administrators will tell me
01:10:13that,
01:10:14you know, we see that
01:10:15your pataras are going home
01:10:16now at four o'clock as
01:10:17opposed to six or seven
01:10:18o'clock. And so, you know,
01:10:21and so maybe we should
01:10:22hire less, etcetera. So that's
01:10:23the the one thing. And
01:10:24then in terms of
01:10:25people being compensated to work
01:10:27from home, no. It's the
01:10:28same. I mean, people have
01:10:29raised the point that, well,
01:10:30you know, they don't pay
01:10:31for parking and things like
01:10:32that, which is true. But
01:10:35we have not changed compensation,
01:10:37for people that work from
01:10:38home. Oh, you mentioned I
01:10:39think I heard it. Just
01:10:40double check. So you do
01:10:42pay
01:10:45No. I told him not
01:10:46Internet access. So he asked
01:10:48for Internet access. I said
01:10:49I would pay for his
01:10:50license, for him to maintain
01:10:52CME. I'd pay for all
01:10:53the equipment,
01:10:54but I'm not paying for
01:10:55his Netflix Internet access. Yeah.
01:11:02Yeah. Because otherwise, he's gonna
01:11:03want fiber optic and everything.
01:11:05Yeah.
01:11:06Alright. Any other questions?
01:11:08Well, if anyone has you
01:11:09have a question? Uh-huh.
01:11:11Oh, yeah.
01:11:12Without any reimbursement for the
01:11:13new technology,
01:11:16are we
01:11:17foresee maintaining this
01:11:19and figure it financially
01:11:21viable?
01:11:24Yeah. So that's a million
01:11:25dollar question,
01:11:26and I cannot just answer
01:11:27it in a few minutes.
01:11:29But,
01:11:32there there are many ways
01:11:33you can, provide that answer.
01:11:35First, they are digital they
01:11:36are CPT codes for digital
01:11:37pathology. There are thirteen CPT
01:11:39codes. They're tracking codes for
01:11:41now.
01:11:41Trying to get your LIS
01:11:42system to do that is
01:11:43not easy. So if you
01:11:45track those CPT codes, you
01:11:47can show,
01:11:48your administrators
01:11:50how much you digitize and
01:11:51the value it brings.
01:11:53Two is,
01:11:55there's the quality workload factor.
01:11:58So
01:11:59I would, for example, have
01:12:00to send out all those
01:12:02perinatal placenta cases,
01:12:04and we've had to, you
01:12:06know, use that number, which
01:12:07is a huge number, actually,
01:12:10even for perinatal autopsies. For
01:12:11someone to do a phytopsy,
01:12:13that's a four thousand dollar
01:12:14charge for us.
01:12:16So you, you know, you
01:12:17can factor that in. I've
01:12:18also shown them how much
01:12:20it costs to archive. Because
01:12:21when you archive and you
01:12:22request a case from the
01:12:23archive, you have to pay
01:12:24for that. So there are
01:12:26many, many things you can
01:12:27do that builds up. There
01:12:28isn't one
01:12:29ROU
01:12:30ROI. You need many different
01:12:32things to build up an
01:12:33ROI. But the and and
01:12:34I've done all of that
01:12:35and explained it to our
01:12:36institution because I'm asking them
01:12:38to ramp up our digital
01:12:39pathology operation. But at the
01:12:40end of the day, they
01:12:41always ask the same question
01:12:42is, well, how many
01:12:44fewer doctors can you hire
01:12:46through the system? And I've
01:12:47told them the computers are
01:12:48not yet to replace the
01:12:49doctors.
01:12:51And I've told them that
01:12:52it's the same when in
01:12:54two thousand and ten, when
01:12:54we barcoded our lab, they
01:12:56also asked me and Anil
01:12:57Purwani back then, why are
01:12:59we paying for this? Where's
01:13:00the return on investment?
01:13:02Like, where does the barcoding
01:13:04make money? It made no
01:13:05sense to them, to the
01:13:06CFOs and the accountants.
01:13:10And all we could tell
01:13:11them was that, well, this
01:13:12makes us
01:13:13better,
01:13:14safer.
01:13:15You know, we're avoiding mistakes.
01:13:17We're avoiding, you know, mix
01:13:18ups of,
01:13:20cases.
01:13:21We are saving potentially millions
01:13:23in lawsuits, and it was
01:13:24all potential cost savings. But,
01:13:26eventually, they allowed us to
01:13:27do it. And now can
01:13:28you imagine if they did
01:13:29not agree to pay for
01:13:31that infrastructure?
01:13:32But I think they realize
01:13:34that
01:13:35if you don't go digital,
01:13:36you cannot do AI. And
01:13:37everyone's doing AI now, not
01:13:39just pathology. Our cardiologists are
01:13:41using AI. Even our neurosurgeons
01:13:43are using AI to guide
01:13:44them,
01:13:45when they're doing surgery. Our
01:13:46radiologists have, like, seventeen AI
01:13:48algorithms running. So we're actually
01:13:50behind. We have six AI
01:13:52algorithms. We are behind other
01:13:53departments. So they realize that
01:13:55we will not be able
01:13:56to even catch up with
01:13:57radiology, cardiology, some of the
01:13:59surgeries
01:14:00if they don't allow us
01:14:01to go digital.
01:14:03And they know that for
01:14:04a hospital of forty and
01:14:05now we're gonna be forty
01:14:06five hospitals, there's no way
01:14:08I can efficiently run that
01:14:09system. What they really want
01:14:11me to do is consolidate.
01:14:12They do not want forty
01:14:13five histology labs. But if
01:14:15I can have four or
01:14:16five histology labs throughout a
01:14:18whole region,
01:14:20that saves the system a
01:14:21whole lot of money. So
01:14:23they've understood that message. It's
01:14:25hard to tell that story,
01:14:26but you have to tell
01:14:27that story. Hopefully, that answers
01:14:30some of the things we've
01:14:30done.
01:14:32So if anyone else has
01:14:33questions, you can just ask
01:14:34doctor Daisy for my email.
01:14:37She'll screen it first, and
01:14:38then you can, she'll send
01:14:39it on to me.
01:14:42Okay. Thanks for your time.