Latest / Show Me The Evidence / From Physics-Based Simulation to Surgical Robots| Show Me The Evidence E8- Dr Dwight Meglan

Transcript

From Physics-Based Simulation to Surgical Robots| Show Me The Evidence E8- Dr Dwight Meglan · Show Me The Evidence

0:07
In the late 1990s, Dr. Dwight Meglan and I met at
0:12
Medicine Meets Virtual Reality
0:14
somewhere in California, I seem to recall.
0:17
Dwight's an engineer. Initially, when I met him, he was
0:21
working on physics-based virtual reality simulation.
0:24
He subsequently moved into surgical robotics.
0:28
You're very welcome this evening,
0:31
Dwight, and nice to talk to you.
0:34
So you developed one of the first
0:37
physics-based virtual reality simulators.
0:39
I purchased one in 1999 at some considerable cost.
0:44
So at that time, how did you envision that the VR
0:49
simulation market would develop?
0:52
And has it developed the way that you thought it would?
0:57
I have to be honest, I really didn't think about it a whole
0:59
lot about relative to a market.
1:01
I was more thinking about it as a technical problem.
1:05
I was just interested in the idea of how do you do
1:08
real-time physics of tool tissue
1:10
interaction and replicate the real-time fluoroscopy
1:15
and give haptic feedback in a way
1:17
that was natural to the procedure itself.
1:20
And I was much more focused on that than anything else.
1:25
And I really didn't give it a lot of thought other than
1:27
there's a lot of catheterization.
1:29
So I figured it was probably going
1:30
to be useful and people would use it.
1:34
And do you think that the current developers of virtual
1:37
reality simulation pays much attention to the physics-based
1:42
virtual reality simulation
1:44
today, as you did back in the 1990s?
1:48
I don't really think many of the simulators pay much
1:51
attention to that, no matter what it is.
1:54
I was kind of obsessed with trying to mimic the actual
1:59
physics of the tool tissue interaction.
2:02
And I figured if I could get that right, then the
2:03
experience for the user would
2:05
reflect reality to some degree.
2:08
So I kind of focused on that. Whereas, as you well know, a
2:12
lot of the simulators kind of
2:14
focus on the user experience itself.
2:17
I actually didn't really think about the user experience.
2:19
I just figured if I did the simulation
2:21
right, then the user experience would be right.
2:24
So what do you think have been the drivers of the current
2:28
state of virtual reality simulation today?
2:31
Well, over the last 20 years, 20-odd years?
2:36
Basically, the device manufacturers. I mean, I figured
2:39
early on, I didn't really give
2:41
that a lot of thought, to be honest.
2:42
I was just thinking about it relative to build something
2:46
that's useful for the clinicians.
2:48
And that would be something that would be meaningful.
2:52
But as I spent more time working in the area, because I
2:55
worked on other simulators besides endovascular,
2:58
I came to understand that really the only people that are
3:01
really paying for this and really caring about it a whole
3:03
lot were the device manufacturers.
3:07
And they kind of drive the way things work. And that's kind
3:12
of what it seems like it is today to me.
3:14
I mean, there's various professional societies that have
3:18
set up standards for learning and other things.
3:23
But as far as I know, like the thing I was driving towards
3:26
years ago, like 20, 25 years ago,
3:29
was trying to figure out how to come up with physics tests
3:33
that would allow you to verify
3:35
the correctness of the simulator.
3:38
And those still don't exist for the most part, as far as I
3:41
know, which I'm kind of surprised at.
3:44
Yeah, but one of the things that always struck me about the
3:47
sensation that the or the perceptions
3:50
are that the user actually had from the simulation. People
3:55
have different sensory thresholds.
3:57
Therefore, the sensory thresholds are
3:59
going to impact on what they perceive.
4:01
And one of the one of the interesting things I find out
4:03
over the years was the plastic surgeons or microsurgeons.
4:07
They won't go for a run. They won't have coffee.
4:10
They try to get a good night's sleep the
4:12
night before they operate because they know
4:14
that they can, a slight nudge can
4:17
impact their performance the following day.
4:20
So it's very hard to actually come up with, you know, what
4:23
are the sensory thresholds for the clinicians to feel?
4:26
So how did you approach that or try and solve it?
4:29
Well, so we did do a couple of things early on because we
4:32
were kind of curious about how would we
4:35
know if we were producing
4:37
things that the cardiologists felt.
4:43
And what we ended up doing was we built an instrumented
4:46
torquer that the cardiologists would
4:49
hold to actually manipulate the tools.
4:51
And we figured if we could measure what
4:54
they were actually feeling in real life,
4:58
then we could actually, and we knew what we were feeding
5:00
back to them on the simulator,
5:01
that if they were at least somewhat correlated, that we
5:04
were actually doing at least some level of a good job.
5:07
We never went far enough along with that to actually try
5:10
and figure out if there's
5:12
different thresholds for different people,
5:13
which of course there is. We know from haptics that
5:16
different people have just
5:17
noticeable difference thresholds.
5:20
But we never pursued that. I mean,
5:23
this was never done academically.
5:25
This was all done in the auspices of a company.
5:29
And since I'm not super active in the field anymore, like
5:35
you said earlier, I mostly work on surgical robots now.
5:38
I'm not aware of anybody having published extensive data
5:42
sets on what it is that a
5:43
cardiologist feels when they do catheterisation.
5:46
So there's not minimal data out there
5:49
to even test against, quite frankly.
5:51
And we certainly don't know much about the actual
5:54
interaction of the catheters with the tissue,
5:58
which is what I was really headed toward. I really wanted
6:00
to eventually be able to say that
6:03
we're actually generating the same forces, contact forces
6:07
and pressures and motions that
6:09
you see in real life quantitatively.
6:11
Yeah, I think that's still a real problem today.
6:15
Sometimes I had four of the simulators that you built when
6:20
I worked in University College Cork.
6:22
And occasionally the haptics weren't switched on and the
6:27
clinicians would be telling me
6:28
only the haptics in this are great.
6:30
And I'm not sure what's going on
6:33
here. I mean, I know about that.
6:38
I think over the years, I think it's just a personal
6:43
conclusion that I've come to.
6:44
Is it the better the clinician is, the
6:46
more attentional capacity that they have,
6:49
the greater awareness they have to the
6:52
subtle aspects of catheter wire performance?
6:56
And I think it's the same in
6:59
robotics or actually there's no.
7:06
Well, except for in the DV5 now does have it.
7:10
So, and I've never used it, so I don't know what it's like.
7:13
So.
7:16
So you think one of the biggest drivers
7:19
of the physics based virtual reality simulation
7:22
was primarily the manufacturers.
7:24
I mean, and the conclusion or the observation
7:27
I've sort of made is they seem to want it only
7:30
for mostly for marketing and that education and training
7:35
seem to come well down the track after that.
7:38
Would you agree with that?
7:40
I mean, everything's subservient to a purpose
7:44
when it's a product, right?
7:46
I mean, they're trying to sell stuff
7:48
and that's just the way it is.
7:50
They'd like you to use it well,
7:51
or at the very least they'd like you
7:53
to not use it dangerously,
7:54
because if nothing else, it's bad on their reputation
7:57
and maybe it can be more than that,
7:59
depending on the situation.
8:00
Could get them into liability issues or whatever.
8:03
So they want you to use it in a way that
8:06
they consider you to be competent at running the device.
8:11
But I mean, I honestly, having worked on the simulators
8:16
and on the robots, I mean, there's a threshold
8:20
at which they kind of draw the line.
8:23
They just want you to be able to use the tool
8:26
in the way that they had designed it to be used,
8:28
at least to their best understanding
8:29
of how it designed it to be used,
8:32
but not, they don't, as you know,
8:35
they don't claim any responsibility
8:37
for the practice of medicine.
8:38
So they're not trying to teach you how to do the procedure.
8:41
They're just trying to teach you how to use the device.
8:44
So that's the cutoff.
8:46
I'm not entirely sure, but I had a conversation
8:49
with a manufacturer last week,
8:52
and it was about the use of a stapling device.
8:56
Okay.
8:57
And the stapling device has a six to 27%
9:00
leak and complication rate.
9:02
And one third of the patients that develop a leak will die.
9:07
That's huge consequences for a relatively--
9:10
And the conversation that we're having now saying that
9:13
basically what you want to be able to do is,
9:17
now, well, number one, we have good evidence
9:19
that a lot of the very experienced surgeons are,
9:23
I don't know that they're not able to,
9:24
but they certainly don't know how to use the device safely.
9:29
Okay?
9:29
And to what I said to the manufacturer,
9:31
I said, you really want to be able to train the surgeons
9:36
to use your device safely according to the IFU.
9:41
And what they were concerned about was
9:43
that what I was proposing was that they were training
9:46
surgical skills and I wasn't.
9:48
I was just arguing that they should be training
9:50
to use the device safely.
9:53
Yeah, I would agree with that,
9:55
with what you just said on,
9:56
I mean, this is all just about doing the stapling itself.
9:59
It's not about doing the procedure.
10:00
And I mean, that's what the stapler is, right?
10:03
So yeah, I can see that.
10:05
But I would have thought that would be
10:07
that the manufacturer would take that in
10:09
as that is their responsibility,
10:11
is to train them how to use the device effectively,
10:15
which basically means safely when you can write down to it.
10:18
That's correct.
10:19
I mean, and a lot of the problems that we see
10:21
with new devices or even old devices,
10:25
the clinicians will blame the device
10:28
but say that if something was wrong,
10:29
it was the device fault.
10:31
And I mean, for the cardiovascular devices,
10:33
the implanting, the cardiac defibrillator.
10:40
See, clinicians, they don't follow the instruction for use.
10:45
For example, when you're suturing the device
10:49
into the packet that you've created,
10:51
some clinicians don't use any sutures at all.
10:53
They don't use the sleeve that covers the leads.
10:57
They use the wrong type of sutures.
10:59
They use to dispose or sorry, sutures that melt away
11:03
rather than, you know,
11:05
For a minute.
11:06
Surely I would have thought that the manufacturers
11:08
would be concerned about that.
11:11
I would think so.
11:12
Yeah, but yes, they're concerned about it.
11:15
And their trainers were telling them
11:18
that maybe some of the clinicians
11:20
that were training the juniors
11:23
were not maybe as good as they originally perceived.
11:27
You know, they had a senior position.
11:29
They had a high volume of procedures per year
11:32
according to their CV.
11:35
But when you looked at their performance,
11:37
they really didn't do that well.
11:39
And over the last decade, what we've been publishing,
11:42
you know, for the concert validity of the metrics
11:44
for the different devices,
11:46
what we've been observing is
11:47
that some very senior clinicians
11:49
perform worse than the worst trainee.
11:53
Now that's problematic.
11:56
Yeah, no, that's not good.
11:59
Yeah, it's, you know, the one question there is,
12:03
you know, are they really that bad
12:05
or is it because they're not acclimated
12:08
to the device and the technology never have been,
12:11
you never learned how to use it well.
12:13
This wasn't on the device.
12:14
This was actually video recording from actual patients.
12:18
Okay, just, okay.
12:20
Yeah, and the metrics weren't made by,
12:23
us, they were made by their peers
12:25
and for a straightforward procedure,
12:27
not for a complex procedure.
12:29
Yeah, yeah, well, these, as you know,
12:33
when you're doing things like this,
12:34
we were talking about,
12:35
these are all gonna be observational metrics, right?
12:38
Because it's surgeons watching other surgeons
12:40
so they don't have any other thing they can do
12:43
other than watch and say what's going on.
12:45
Whereas, you know, like in a simulator,
12:47
you could actually measure the actual forces
12:50
and, you know, things like that,
12:52
you know, like tool tissue interactions and stuff
12:55
and turn those into metrics if you wanted to,
12:57
but then nobody would know what those metrics need to be
12:59
because we don't know what that data is in real life.
13:02
So we're kind of stuck doing the same thing,
13:04
just making it fancy and quantitative
13:07
in terms of like tools moving around and stuff.
13:10
You know, all the economy, emotional,
13:12
the other stuff.
13:13
We'll get to that one another.
13:15
So when you were building the simulator in the 1990s,
13:19
what was your view on what the simulator should measure?
13:25
And-
13:26
Basically everything.
13:29
There's only one thing worse than having no data.
13:31
That's having too much data.
13:32
It's too much.
13:35
So measuring everything.
13:37
Well, no, no, no, no, there's a reason for that.
13:41
So when I did the original VIST,
13:44
I did try to do a harness on it
13:48
that would record literally everything,
13:51
like every force, every torque, you know,
13:53
every contact point, like everything,
13:56
because I figured eventually we were gonna wanna go through
13:59
there and try and make sense of it
14:00
in terms of what would give us measures
14:02
that would be kind of telling you
14:04
whether you did it right, whatever it was.
14:08
And so the original intent was literally to record
14:11
as many things as we could
14:12
and then go rifling through it
14:14
and try and figure out what the essential measurements
14:17
would be that would tell you
14:19
whether you had actually learned how.
14:21
The, so the simulators were originally supposed
14:24
to be designed around learning goals.
14:27
And we wanna be able to have a way to actually
14:29
a certain that you had learned whatever the goal was.
14:32
And so our intent was that we would have some kind
14:36
of measurement of that.
14:37
And since we didn't know what that would be,
14:40
we were like, let's just do built-in suspenders
14:42
and record everything and then we'll figure out
14:43
how to pare it down to something that makes sense.
14:47
Well, to some extent, we did that
14:48
for the mechanical thrombectomy for acute stroke.
14:53
I mean, we developed the metrics from the human procedure
14:57
along with Mentice.
14:58
And then what we did was,
15:01
myself and the three interventional neuro radiologists,
15:04
we imposed the metrics well, Mentice did.
15:07
We told them what to do
15:08
and they imposed the metrics on the system.
15:11
And one of the things that the simulator didn't do
15:14
was when you inject contrast into the brain
15:17
and into the brain neurovascular structure
15:22
through a microcatheter, they were using syringe with air.
15:26
And it just plunged straight down and--
15:30
Wrong flow characteristics.
15:32
Absolutely.
15:33
And the interventional neuro radiologist stood the ground.
15:36
They basically said, that's not gonna do.
15:37
Yeah. And we're not standing over this.
15:39
So Mentice had to develop and subsequently patent
15:43
a new methodology where they were injecting a fluid
15:46
into a small device that actually was made
15:50
the whole training more realistic.
15:53
And I mean, it works when you insist on it,
15:57
but you have to start off with the metrics.
15:59
So you were trying to measure everything
16:01
in the late 1990s, early 2000s.
16:06
What did you end up,
16:07
what sort of metrics did you end up producing?
16:09
Or the simulator?
16:10
Well, so I really was not the one that did that.
16:16
That was others that decided what the metrics would be.
16:19
I just put the platform together
16:21
so that the information would be available.
16:23
And then other people made decisions about
16:25
what the actual metrics derived
16:30
from that information would be.
16:32
These were clinicians or clinician engineers or engineers?
16:37
Yeah, my experience was that
16:39
it was a little bit of everything.
16:40
It was clinicians in concert with the companies.
16:44
The companies usually worked with some clinicians
16:46
and it was usually a small number.
16:49
And whoever their subject matter experts was
16:53
that they were paying to consult with them.
16:57
And then they would basically for the most part
16:59
do whatever their SMEs said.
17:01
And all we did was implement it.
17:04
Which usually meant just pulling out things
17:06
that were already there.
17:08
It wasn't that often we had to do something new.
17:11
So it was a combination of stuff
17:13
that was already getting dumped out
17:14
or that was already available in the simulator.
17:17
And then turning those into some kind of metrics
17:19
which were typically, as you know,
17:22
they're observational metrics.
17:23
They're nothing exotic for the most part.
17:27
And
17:28
I know that you keep an eye on the simulation developments.
17:35
What's your view on the metrics that currently exist?
17:38
Not just in endovascular simulators or
17:41
interventional cardiology simulators,
17:42
but robotics and so on.
17:45
They haven't changed a lot, basically.
17:48
They're kind of the same things
17:49
we came up at the very beginning
17:50
when you come right down to it.
17:52
I mean, I think it's slightly worse.
17:53
Yeah, I'm sure there's...
17:56
Okay, you're much more in the
17:57
fray than I am these days, frankly.
18:00
I was at the ERA conference in Paris a few months ago,
18:04
and I walked around all the
18:05
booths to look at the simulations.
18:08
And almost none of them, I think only one of them,
18:11
had any form of metrics whatsoever.
18:14
So it was...
18:15
I mean, I was...
18:17
To be honest with you, I was kind of shocked.
18:19
And...
18:20
Well, were they just using them as a marketing tool?
18:22
I mean, a lot of the early
18:24
simulators we did were just marketing tools.
18:26
You know, you stuck them in a booth
18:28
and it was a way to show off the system,
18:30
you know, whatever the system was you were trying to sell.
18:32
And they really didn't care if there were metrics or not.
18:34
It was just to give people an experience.
18:36
Yeah. Well, I also talked to some of the manufacturers,
18:40
the device manufacturers, and they
18:41
were complaining about the simulators
18:43
and the fact that they weren't able...
18:45
Well, some of the manufacturers, device manufacturers,
18:48
didn't want to measure things.
18:50
Because...
18:51
Because if they...
18:52
They don't want to know.
18:54
Correct. They don't want to know.
18:55
Plausible deniability.
18:57
Plausible deniability.
18:59
And other manufacturers complained that the simulation
19:04
companies weren't producing the
19:06
metrics that they required. And they said, where do you
19:09
expect them to get the metrics from?
19:11
You really have to derive the metrics from real clinicians
19:15
doing real procedures in real life
19:19
situations. And a straightforward procedure, not a complex
19:21
procedure. And when they go to the
19:23
manufacturer, they say, these are our metrics, our
19:27
validated metrics. If you can build the
19:31
metrics into your simulator, we'll purchase it. If you
19:33
can't, we're not. So I mean, go on.
19:38
That would require people to actually
19:39
say that, that that's what they want to do.
19:42
You know, if people still buy the
19:44
simulators, even if their metrics aren't there,
19:47
you know, what am I supposed to say? Yeah, well, I think
19:51
something's got to give, because
19:53
what we're seeing is that the evidence is fairly clear now.
19:58
The intraoperative performance of the
20:00
clinician is going to impact on the patient outcomes. I
20:03
mean, the data is there now. And
20:07
with a new device, it's more likely that a clinician who's
20:12
not trained, you know, to a level of
20:15
proficiency, they're going to perform less well than
20:18
somebody who's actually been trained to
20:19
efficiency. And that's the data on that's very clear. So I
20:23
don't understand why there isn't,
20:27
you know, a drive by the manufacturer, the device
20:29
manufacturers to insist on metrics being
20:33
implemented in the simulator. I don't know what to tell
20:40
you. I mean, maybe they don't think that
20:42
it's going to doesn't make a difference on selling their
20:45
gadgets in any form. It's just a cost,
20:49
to do this. And so if it's not going to make them sell
20:53
more, why do it? I don't know.
20:55
Well, I mean, the other side of that is there's devices
20:59
that have failed in the marketplace.
21:01
And it wasn't an issue with the device. It was the user of
21:06
the device. It's been held back
21:10
and repeated the introduction of that device into the
21:13
market. And I've been told that it's
21:16
hundreds and hundreds of millions in some cases. Yeah, I
21:20
think a lot of times people don't think
21:22
ahead of like think ahead like that. Or those or those
21:25
stories don't get around well enough,
21:27
right to management at the medical device companies to
21:31
realise, really, you know, you could. Yeah.
21:36
Hmm. So in the early 2000s, she appeared to move away from
21:43
simulation and more into surgical
21:44
robotics. What were the major drivers of that change for
21:48
you? So I basically kind of felt
21:52
hemmed in with surgical simulation.
21:55
You were always constrained by,
21:58
there's not that many things to be simulated,
22:00
at least at that point.
22:02
And it was always driven by the device manufacturers,
22:06
whatever it was that we were gonna be doing.
22:08
And quite frankly, I kind of felt like
22:10
it was kind of a lousy business
22:12
because you were inherently capped
22:14
as to how many things you were gonna be able to sell,
22:16
how much money you were gonna be able to make.
22:18
Not that I really cared that much about it,
22:21
but it just didn't seem like it was a major growth field.
22:24
And surgical robotics had really just started to get going.
22:28
And it was interesting,
22:30
and there was a whole bunch of different things
22:32
to be done there.
22:34
So I just decided to move over there.
22:37
And as it turns out that surgical robotics
22:39
isn't all that different from surgical simulation
22:41
in terms of sort of the systems engineering stuff,
22:44
which is more where I work at it.
22:45
So I like working at the real-time sensing,
22:49
real-time computation, real-time actuation,
22:52
and then potentially real-time graphics,
22:55
whether it's stereo or whatever.
22:58
And simulators are actually very similar to robots
23:00
as far as that goes in terms of all the different parts
23:03
having to go together and work in real-time.
23:05
So I basically had a skillset already done
23:09
because of the simulators that moved over onto robots.
23:12
And so I just started working on the robots
23:15
and kind of been going ever since.
23:19
So you agree with Rick Satavo then
23:20
that the VR simulators and surgical robotics
23:24
have an analogous DNA, if you like,
23:27
as they're both information processing systems?
23:31
Yeah, they're both interesting.
23:35
They're both systems where you're doing real-time sensing
23:38
and computation and actuation
23:40
and typically graphics generation.
23:43
The simulators don't nearly have as high a bar to them
23:47
as the robots do.
23:48
I mean, the robot can actually hurt somebody.
23:50
The simulator can't directly hurt somebody.
23:52
It could indirectly hurt somebody by teaching them wrong,
23:55
but it's not got a direct immediate effect
23:57
of punching a hole in someone.
23:59
So the robots have a much higher standard
24:01
and there's much more that has to be done
24:03
with verification.
24:04
In fact, robots do all kinds of things with verification,
24:07
whereas the simulators basically don't.
24:10
What do you mean they do all sorts of things
24:12
with verification?
24:13
You mean within the system
24:14
or within the management of the system
24:16
by the manufacturer?
24:19
When you're developing them,
24:21
they put together all kinds of bench tests for the robots
24:24
to make sure that they have the right forces,
24:27
the right velocities, the right accelerations
24:30
that they hold up over time, your reliability,
24:35
that they're stable, that the control loops are stable,
24:40
all kinds of things like that.
24:42
There's just dozens and dozens of tests
24:45
that the device manufacturers put together
24:48
to verify that the specifications
24:51
that they've made for the robot are what the robot does.
24:55
And I, at least in my experience,
24:57
there's little to none of that
24:58
that happens with the surgical simulators.
25:00
It's, if anything, it's validation,
25:03
which I basically don't like validation.
25:07
Verification is like you did the engineering correctly.
25:11
Validation is like, "I think it's okay."
25:14
It's not precise for me.
25:16
That's like the end user just saying,
25:19
"Okay, that worked, I like it."
25:21
But they don't know why they like it or don't like it.
25:23
I'm not sure I agree with that.
25:25
I mean, validation-
25:26
This is from an engineering point.
25:28
This has been an engineering point of view.
25:30
Clinically, I mean, it's not my methodology.
25:34
The methodology was agreed in the 1970s
25:37
by all the education and training establishments
25:40
in terms of how you validate,
25:43
whether something works or it doesn't work.
25:45
Well, okay, so from an engineering point of view,
25:47
validation is like, "Yeah, okay, it does something useful.
25:53
They're actually learning or whatever it is."
25:55
But I find that to be imprecise.
25:59
From an engineering point of view,
26:01
I'm much more interested in verification.
26:03
I guess the way I look at it from a,
26:05
like from a surgical robot point of view,
26:08
verification and validation is,
26:10
verification is I built what I said I was gonna build,
26:12
and validation is I built the right thing.
26:15
I consider that much less precise
26:19
from an engineering point of view.
26:21
Yeah, but it isn't engineering.
26:23
And I mean, the ultimate test of the validation
26:26
for a robot or a simulator is,
26:28
does the robot do what it's supposed to do clinically?
26:33
For somebody that can it do the procedure,
26:36
can it do the procedure safely?
26:38
And that's not the engineer's realm, basically.
26:42
That's my point of view.
26:43
That's for the clinicians and others,
26:47
but the engineer is to verify.
26:49
Yeah, so in the 1990s, laparoscopic surgery developed,
26:56
they extensively investigated this whole surgical tracking
27:01
and path and path length and motion and so on.
27:05
And it really didn't go anywhere.
27:08
And surgical robotics seems to be heading down
27:10
the same road as an approach to the assessment
27:15
of operative performance.
27:19
Sure.
27:20
Oh, I think there's two parts to this.
27:23
It's basically the only thing people can think of
27:25
for the most part.
27:26
And it's gonna get turbocharged
27:28
because of machine learning in AI.
27:30
Yeah.
27:31
It's gonna be on steroids
27:32
because they're gonna be looking for patterns
27:33
in everything with the data they collect.
27:36
I have a feeling.
27:38
Yeah, but I mean,
27:39
I mean, it's like looking for a target
27:44
and a very thick fog.
27:47
I mean, the whole thing about science
27:49
is that you're supposed to hypothesise
27:52
what it is that you're looking.
27:53
I mean, the tracking stuff,
27:58
it seems to be like going on a fishing expedition
28:02
without any--
28:02
It is, yep.
28:04
I mean, and do you think it's gonna go anywhere?
28:08
I'm sure they'll find some things.
28:11
I mean, this is the thing with AI coming at it.
28:15
It's that it's able to do so much more
28:18
than we could in the past.
28:22
But I mean, you really wanna have a theory of operation
28:26
behind things of what is it we're trying to achieve
28:29
and how is it you're going about
28:32
making this thing attain whatever it is
28:34
it's going to attain.
28:36
So I would think that you basically wanna,
28:42
you have to have some kind of concept
28:44
that drives why you do what you do,
28:46
which is why you do the things you do
28:48
where you break procedures down
28:49
so you understand that there's learning goals
28:52
and based on those learning goals
28:53
that there's a scaffolding of skills that support those.
28:57
And then from that,
28:58
you'd figure out how to measure things
29:00
because those are the things that support
29:02
your understanding of whether you would obtain
29:04
those skills or not.
29:06
Yeah, so what strikes me is that what laparoscopic did
29:09
and what robotic surgery is doing now
29:12
and interventional cardiology and the fluoroscopically,
29:15
they measure process and they take that
29:19
as some sort of indication of skills.
29:22
If somebody does lots of procedures,
29:24
they're gonna be better than somebody that does less.
29:26
If you're smooth with the instrument
29:28
or you have a shorter path length,
29:30
you're obviously more skilled
29:31
rather than you just jump complete phases
29:34
or steps in the procedure.
29:36
And this process seems to be,
29:40
I mean, like they're hanging on to,
29:41
this is a strategy for measurement.
29:43
Yeah, well, so if it was a perfect world,
29:47
I'll just use suturing as an example,
29:48
because that's the thing I tried years ago.
29:51
If it was a perfect world,
29:52
we'd actually know exactly how much tissue apposition,
29:56
how much of the tissue faces need to come be in contact
29:59
and how much pressure there needs to be,
30:02
and how well distributed it is.
30:04
And that would actually be the measure
30:05
of how well you did suturing.
30:08
It wouldn't be an indirect measure of like,
30:10
did you do it smoothly, quickly, decisively,
30:14
all those things,
30:15
because that's not really the measure of the outcome.
30:18
We know that the whole idea with suturing
30:20
is you're trying to get the tissue
30:21
to grow back together and heal, right?
30:23
So you'd measure it based on the metrics
30:25
that are reflective of achieving
30:27
the ineffective environment for tissue regrowth.
30:32
But we don't even know what those are.
30:35
Back in the day, at least when I was working on it,
30:37
that wasn't published information.
30:39
Nobody had sat down and built like a jig
30:42
and tried all kinds of different amounts of pressure
30:44
and distribution of the sutures along it
30:48
and the likes to see how well living tissue would regrow.
30:52
So like, nobody knows as far as I know.
30:55
Yeah, well, I mean, I've worked quite a bit
30:57
with suturing as well,
30:58
and I haven't seen once a milliter
31:00
that can reasonably, even reasonably,
31:04
approximate the instrument thread, needle, suture,
31:09
or tissue dynamics.
31:10
And so what I've reverted to
31:12
is actually getting individuals to suture on tissue
31:16
and then a surrogate measures,
31:18
like how well is the tissue approximated,
31:21
did the surgeon drive the needle
31:23
through the tissue atraumatically,
31:26
do their knots hold when they're supposed to?
31:29
Is the tissue pucker?
31:30
And those surrogate measures seem to work reasonably well,
31:34
but why hasn't somebody actually worked out
31:37
the actual physics of suturing and thread dynamics?
31:44
Well, the suture and the tissue, it's not easy.
31:49
It's a high computation activity
31:53
and there frankly aren't that many people
31:56
that work in this field.
31:58
If this is, so like look at what's happening
32:01
with AI right now.
32:02
The entire planet has decided that having physics,
32:07
what they're calling physical intelligence now
32:09
is sort of one of the next big things for AI.
32:12
So there's like thousands, tens of thousands of people
32:15
doing physics models now for robots,
32:18
for humanoid robots and everything else.
32:20
And you're gonna see a tremendous amount of improvement
32:23
in the physics simulation abilities.
32:25
You just don't have that with surgical simulation.
32:28
It's not like the whole planet's decided
32:30
this is how you're gonna become a billionaire
32:32
is by getting physical intelligence
32:34
combined with humanoid robots,
32:36
which is what a whole lot of people have decided now.
32:39
So you're seeing a huge amount of improvement.
32:40
You're just not gonna see that in simulation,
32:42
in surgical simulation.
32:44
It's just not, like I said,
32:46
that's the reason I got out of it a long time ago.
32:48
It's just not a big market.
32:49
So there's not a lot of drive to do it.
32:51
What's gonna happen with surgical simulation
32:53
is they're gonna get the indirect benefits
32:56
of all the work that's happening for robotics right now.
33:00
And you have all these things that people are doing
33:03
where they're using AI combined with physics models now,
33:07
so that you can use a reduced accuracy physics model
33:11
and you use AI to fill in the blanks.
33:14
I think you'll probably see that showing up
33:16
in surgical simulation,
33:18
but it's gonna come from other fields.
33:19
It's not gonna come from surgical.
33:21
It's not gonna be developed in surgical simulation.
33:23
It's gonna be developed in other fields
33:25
and adopted in basically.
33:27
But if the measurements of whatever task
33:31
you're at let's stick with suturing,
33:34
if the measurements that you have are measures of process
33:39
rather than actual measures of skill,
33:41
I'm not too sure how the AI system's gonna improve that.
33:44
No, yeah, no, no, yeah, no, that doesn't,
33:48
that's such an indirect measure.
33:50
I don't see how that really is a good reflection
33:52
of mastery, if you will.
33:55
Yeah, I mean, so you're still going to need
33:58
the performance metrics which are derived
34:00
from experienced clinicians
34:03
and to actually teach an AI system to how to assess.
34:08
And I don't see anybody doing that in the surgical domain.
34:12
And I don't know whether they're actually doing it
34:15
in other domains like famous or infamous, whichever.
34:21
Car manufacturer said that surgeons would be replaced
34:23
in what I can remember, was it two years or three years?
34:28
Oh, him, yeah, right.
34:30
So what is happening now in humanoid robotics
34:34
is they're trying to teach all these humanoids.
34:36
So they're measuring all kinds of things
34:38
about everyday activities.
34:41
So it'll be interesting to see if some of that stuff
34:44
actually maybe becomes learning points relative
34:49
to things that could be moved over
34:51
into surgical simulation, just the process
34:55
by which they're doing it, not the details of it.
34:57
Yeah, but you see--
34:59
You are trying to quantify things like crazy right now
35:01
in everyday activities, so.
35:04
Yeah, but what I see a lot of the AI people do
35:08
is they're loading data into the AI system.
35:13
And there's no quality, little or not very good
35:16
quality assurance of the data
35:18
that the AI system's learning on.
35:21
And the individuals that are coding the tasks,
35:26
I mean, they don't really know whether the task
35:30
could or not.
35:31
I mean, but that has to impede the development of AI
35:38
in terms of the assessment of performance.
35:43
Well, in surgery, certainly it is.
35:45
I mean, if you're looking at all these
35:48
humanoid robots desperately trying
35:50
to do everyday activities, I think the thing
35:53
that's happening there is there's so many people doing it
35:55
that it's an aggregate result.
35:59
There's just so many, there's like 100 humanoid
36:02
robotics companies in China alone.
36:05
So you get enough of them doing it
36:07
and they're all piling on top of each other.
36:10
And no matter how much they're screwing up,
36:11
there's enough of them doing the right thing
36:14
that it accumulates.
36:15
And you just don't have that effect with surgery simulation
36:19
because there's just not that many people active in it
36:21
when you come right down to it.
36:23
One of the things that has struck me across the decades
36:27
where we've been hanging out together
36:29
is everybody's looking for an easy way
36:34
to actually do something that's quite difficult.
36:37
And science, I mean, doing good science, it's not easy.
36:43
You know, I mean, and I mean, one of the things
36:47
that I'm shocked at is the use of Likert scales
36:51
for the assessment of performance.
36:52
Because again, it's a measure of process.
36:55
Yeah, yeah.
36:56
I mean, in robotics, I'm sorry, go ahead.
37:00
No, I was gonna say all those hand-wavy things, right?
37:02
All those hand-wavy. One to five, right?
37:05
One to five.
37:06
I mean, my fellows, well, they
37:09
know my views on it, don't they?
37:10
And they initially think that Tony's got a thing
37:14
about Likert scales, and they very quickly learn,
37:17
okay, for the binary metric-based assessment,
37:21
you've got to watch the entire video and score it.
37:23
And you have to learn how to score it reliably
37:26
with somebody else that you're paired with.
37:29
With the Likert scales, you can watch it, you know,
37:32
about two to five minutes of the video,
37:34
and you can score it in one to five
37:35
on the different constructs.
37:37
I mean, that's not measurement, so it's not.
37:40
Yep, yep.
37:42
So what's your current view
37:44
on the status of surgical robotic simulation
37:47
and the metrics used to assess performance?
37:50
I mean, you've worked on these or you've--
37:52
Yeah, I've been around them, yeah.
37:54
You've been around them?
37:57
They could be better.
37:59
I mean, you know, even doing the simple things
38:04
like ring exercises and stuff like that,
38:07
they still haven't verified that the performance of the,
38:11
you know, like the exact forces
38:13
that you grasp the rings with and the exact forces
38:16
when you pull on the stretchy ones,
38:18
that those actually reflect what happens in real life.
38:22
So it's all based on tweaking and tuning
38:25
to make it so that it's at least objectionable
38:29
to the largest number of people,
38:31
but that doesn't mean that it's correct.
38:33
And I don't think it's ever,
38:36
unless they change how they do things,
38:38
it's probably never gonna be,
38:39
you're never gonna know how correct it is
38:40
when you come right down to it.
38:42
But surely that should be of concern
38:45
given that you said earlier
38:47
that the robot is something that can actually hurt
38:51
a patient relatively easily
38:53
if the surgeon doesn't actually know
38:56
how to use it properly and safely.
38:58
Yeah, well, I'm not disagreeing with you on that.
39:02
I mean, but the manufacturer's goal
39:04
is to verify that it won't unintentionally hurt somebody.
39:10
If it intentionally hurts somebody because of the operator,
39:13
that's kind of not their problem,
39:16
is there kind of their point of view?
39:18
Unless of course it happened so many times
39:20
because of something that's clearly
39:22
a human user interface issue,
39:24
then that does become their problem.
39:27
And so whose responsibility is then
39:30
for the clinician to be trained
39:33
to a level that allows them to use the device safely?
39:40
Well, I would assume it would be
39:43
the professional societies.
39:45
I mean, the ones that are actually saying
39:47
that yes, you are a surgeon.
39:49
And if not the societies,
39:51
it's the hospitals themselves giving privileges.
39:55
So then that would be sort of the next level up
39:57
because the privilege granting
40:00
is at a whole different level
40:02
than the device manufacturers, right?
40:05
The device manufacturer is just saying,
40:06
I taught you how to use the device
40:08
in a way that we certify that it can be used.
40:13
But the granting of privileges is saying,
40:15
you know how to use this gadget to actually
40:17
get a clinical result and we're gonna take,
40:21
we end up having liability for it
40:24
if you don't essentially, right?
40:27
But in the US, you have a much better,
40:30
I'm okay, it's problematic,
40:31
but you have a better privileging system
40:33
than we have over in Europe.
40:36
I mean, once you're qualified,
40:37
I mean, I was at a PhD viva by last week
40:40
and one of the examiners was saying
40:42
that he works in Sweden.
40:45
And once he's qualified as a surgical specialty,
40:50
he can do almost anything.
40:51
He also works in New York.
40:54
And he said each procedure that he wants to perform,
40:58
he has to be privileged for it.
41:01
That seems like a better system than here.
41:03
Yeah, yeah, no, I would go with that,
41:06
but I think that's sort of reflective
41:07
of sort of the old school, you know,
41:09
doctors are geniuses kind of thing
41:12
from the old days, right?
41:13
So, yeah, it's not--
41:15
You mean in Europe or the US?
41:19
Well, in Europe in that case, I mean, I don't know if the
41:22
experience you said about in New York, if that's uniform
41:24
across the US, I don't really follow that.
41:27
It would be nice if it was. I suspect
41:29
it's not. It's probably highly variable.
41:31
I can only talk about, you know, I've worked in Yale
41:35
and Emory and it was the case
41:36
in both of those like 20 odd years ago.
41:38
That's good.
41:39
But one of the problems, the Chief of Cardiology, there was
41:42
a credentialing going on for carotid
41:43
artery stenting with embolic protection
41:46
and he was saying that he had to go to one of
41:48
the credentialing committees and I said,
41:51
"That sounds, you know, pretty powerful."
41:53
And he said, "No, not really."
41:55
He says the credentialing system in a hospital is based on
41:59
the person who has the lowest
42:00
experience in that procedure on the committee.
42:04
So there was again no objective for it.
42:08
Nobody wants to have any kind of social admittance of
42:11
social whatever that they don't know what's
42:13
going on.
42:14
Yeah, okay.
42:14
I mean, I think that at some point there needs to be some
42:19
objective assessment of performance,
42:22
either at the end of a course or a
42:24
verification, you know, at a hospital level.
42:29
Because one of the things we're learning is a lot of the
42:32
metrics in the past were used
42:34
for the assessment of performance,
42:35
like how many procedures did you do?
42:37
Did you do a fellowship?
42:38
How long was your fellowship?
42:40
These were the performance metrics that
42:42
were used to say, "Okay, that person's safe to
42:44
do the procedure."
42:45
But one of the things that we've learned over the last few
42:48
decades since the introduction
42:49
of the laparoscopic surgery, to some
42:51
extent, maybe even to a large extent, those sorts
42:54
of metrics mean nothing.
42:56
They're measures of process again.
42:58
Yeah, well, they're social proof, if anything, when you
43:01
come right down to it, right?
43:03
I know this person and I think they're okay.
43:05
And that's kind of the end of it, right?
43:07
Yeah, and do you think that's okay?
43:11
No, of course not.
43:13
I mean, we fall back to what Rick Satava and others went on
43:19
and on and on about years ago,
43:20
about the way that the pilots are done, right?
43:24
Mandatory training and mandatory testing, right?
43:27
Yes, yes.
43:28
And that's that end of story.
43:30
And those simulators, just to throw this in for reference,
43:34
those simulators are verified.
43:37
Those simulators are verified big time, that they give the
43:40
right movements, they give the
43:42
right forces, you know, everything on those simulators is
43:46
reflective of what happens in
43:47
real life.
43:50
But they cost like tens or hundreds of millions to develop?
43:55
Absolutely.
43:56
And so why do you think that the governments and the
43:59
departments of health have not insisted
44:02
on the same level of
44:04
verification for the surgical simulators?
44:08
Two things.
44:09
A surgeon can only kill one person at a time.
44:14
And it's only one person's worth as opposed to hundreds of
44:18
millions of dollars for a plane
44:19
and however many hundreds of people that are going to sue
44:22
the airline to be really blown
44:24
about it.
44:25
Yeah.
44:26
Yeah.
44:26
But I mean, somebody cited in that PhD, that globally,
44:31
there's the number of deaths due
44:35
to surgical misadventure or surgical skills deficits is
44:40
about the equivalent of one full
44:43
jumbo jet going down per day.
44:45
Yeah.
44:46
And that actually doesn't surprise me at all.
44:49
I mean, since air travel has gotten so safe these days.
44:52
But you know, like I said, even if it's that many, it's
44:55
still one surgeon killing one person
44:57
at a time, widely distributed as opposed
45:00
to 500 people in one spectacular event.
45:03
So unfortunately, it just doesn't get in people's brains.
45:07
Yeah.
45:08
I mean, the Bristol case is a fairly famous case in the
45:13
United Kingdom where there was
45:16
a in the Bristol Royal Infirmary, there were two cardiac,
45:21
they were adult cardiac surgeons
45:24
operating on children, children and their
45:27
mortality rate in the Bristol case report
45:31
said that there were there complication, their mortality
45:35
and morbidity was about twice the
45:37
national average.
45:38
I'm told privately that it was
45:39
significantly higher than that.
45:41
That went on for the guts of 10 years.
45:44
And there took 10 years, right?
45:47
Yeah.
45:48
I mean, their peers knew that there were issues.
45:50
And it wasn't until it hit the front pages of Sunday
45:53
newspapers that the colleges of
45:56
surgeons, the Royal College of
45:57
Surgeons was forced to take action.
45:59
You know, that's a very like I
46:02
said, that was that was cumulative.
46:04
And it took a long time.
46:06
And I'm sure that there was a whole lot
46:08
of social stuff going on at the same time.
46:11
Right.
46:11
Like, yeah, these people were known
46:14
to other people and blah, blah, blah.
46:16
And you can't rat them out or
46:18
whatever, whatever it's going to be.
46:20
There's all kinds of stuff like, like I said, back on the
46:22
airplane thing, it's because it
46:24
all happens at once and the stakes are so high.
46:27
If it happens one time, you know, if if a
46:30
surgeon managed to kill dozens of people in
46:33
the span of a week, you know, it
46:35
would become a thing, you know.
46:37
Yeah.
46:38
Yeah.
46:38
There was a wait, wait till tell it wait till tell us.
46:40
Wait till telesurgery goes on steroids and you have
46:43
somebody who's doing way too many
46:46
cases.
46:48
They could make a big mess really quickly, basically.
46:52
What do you mean?
46:53
Telesurgery goes on steroids.
46:54
You mean that there's a lot of it being performed or?
46:56
Yeah.
46:57
And you could have a certain you.
46:58
Well, you and I both know from being in hospitals that, you
47:02
know, the head surgeons bouncing
47:04
between operating rooms with the fellows and the residents
47:07
in there doing doing the most
47:09
of the work, you know, and you places where they have like
47:11
four, five, six operating rooms
47:13
going at the same time.
47:14
But the the lead surgeon is
47:16
supposedly the lead surgeon for all those.
47:19
Imagine what happens when you have telesurgery.
47:22
Yeah.
47:23
I mean, but in the United States, OK, that goes on, but
47:28
it's really not supposed to go
47:29
on the lead surgeon supposed
47:32
to be in the room all the time.
47:35
And they get away with it because they sign off on the case
47:38
or they perform the important
47:39
part of the case.
47:40
You know, I mean, they wouldn't get away with that in the
47:43
UK and Ireland, I don't think.
47:46
Yeah.
47:47
Yeah.
47:47
I mean, there's no two systematic reviews and meta analysis
47:52
on proficiency based progression
47:54
training and comparison to the training outcomes in
47:57
comparison to conventionally trained quality
48:00
assured approaches to training.
48:03
And the results are pretty similar.
48:04
They basically show that if you're trained
48:06
in a proficiency based progression approach,
48:09
you're going to make between 60 and 58 percent fewer
48:13
objectively assessed interoperative
48:14
errors.
48:15
And there's small scale evidence that basically said and
48:19
the differences in clinical outcomes
48:20
is about 60 percent as well.
48:23
So why do you think that the PBP
48:26
approach hasn't been hasn't had wider adoption?
48:31
I mean, this is 20 odd years now, 24 years.
48:36
Well, I would guess, as you well know,
48:39
it's a lot of work to do a curriculum based on
48:41
this.
48:42
I mean, it's a whole lot of work trying
48:44
to figure out what the metrics are and what
48:47
progression is like in an objective
48:49
manner to actually be able to lay that out.
48:52
And it's a lot of work.
48:54
So I think that's a lot of the problem.
48:57
And I think the other thing is that I suspect that people
49:00
don't like the idea of having
49:01
to train to some kind of objective threshold because they
49:05
have no idea how long it's going
49:07
to take them to attain it, frankly.
49:11
But the evidence that evidence is out there now, you can
49:14
train somebody to proficiency
49:16
in one third of the time than you
49:18
can using a conventional approach.
49:20
And it's about less than half of the cost
49:25
to train somebody using the proficiency.
49:27
And I mean, that data is published.
49:30
Yeah.
49:30
OK.
49:31
Well, maybe it's just the words not gotten around.
49:34
This is too new.
49:35
People aren't listening.
49:36
I don't know what to tell you there.
49:40
But it's well, it's not it's too new in the general sense.
49:44
I mean, they've been teaching people the same way for
49:48
hundreds of years, if you will.
49:51
So this is still kind of working its way.
49:54
I don't disagree with you because I'm a big supporter of
49:58
proficiency based progression, as
49:59
you know.
50:01
I think maybe one of the other things is that it needs to
50:03
be made more readily available
50:06
in some kind of an automation way of doing it.
50:09
You know, you and I have talked about that, like how you
50:12
could leverage machine learning
50:13
to actually make it so that you can
50:15
actually do it without having to train people to do
50:19
the evaluations and the likes,
50:21
which is one of the negating factors.
50:23
Right.
50:23
Yeah.
50:24
Well, I mean, that's something
50:26
still in the in the planning process.
50:29
But I mean, I was asked the question in Australia, I was at
50:33
the surgical robotics meeting in
50:37
Melbourne and somebody asked me, Tony, if
50:40
you've got all this, I was talking about
50:41
proficiency based progression training and robotics.
50:44
He said, Tony, if you've got all this
50:45
evidence, why has it not had wider adoption?
50:51
And I hadn't even thought about the question.
50:53
But my answer was out before.
50:55
And clearly, I must have had it in the back of my mind.
50:57
I said, a failure of leadership.
51:00
I mean, the reason that Orsi Academy, the
51:02
reason that the Arthroscopic Association of
51:04
North America and to some extent, now, Medtronic,
51:09
cardiovascular seems to be going down
51:11
that road.
51:12
The reason is somebody senior in the
51:14
organization said, OK, this is what we're doing.
51:18
And this is not negotiable.
51:20
And I mean, but there doesn't seem to be enough.
51:24
ERUS, The European robotic urological society.
51:28
I mean, I showed the data on the failure rate of the ARIS
51:34
fellowship program, the number
51:35
of fellows that weren't finishing.
51:37
And the president, Alberto
51:39
Breda, said, OK, we're changing.
51:42
We're going to go and use a proficiency based progression.
51:44
That's what I call leadership.
51:46
I mean, but it's surgeries, surgery.
51:50
They must have to have more leaders.
51:52
And is it something I'm missing?
51:58
I'm only an engineer as far as that goes.
52:00
Yeah, but you've been around this for a long time,
52:03
and I know that you're using these things.
52:06
Yeah, well, you know, a lot of people,
52:10
yeah, no, I'm just assuming that they
52:14
don't think they need it for whatever reason.
52:18
I mean, you've seen, you know, in a few isolated cases,
52:21
you've seen people make the decision that
52:23
they do need it, and that's that, right?
52:25
I mean, it might also be, I
52:27
mean, like with flight simulation,
52:28
I mean, that's all government regulated.
52:30
It's mandated, you know, and so you'd need
52:33
something like that, but I don't know if that'll ever
52:37
certainly won't happen in the United
52:38
States anytime soon, put it
52:39
mildly. But yeah, I would think it.
52:45
Well, our current environment, which insists that the
52:49
government can't do anything right,
52:50
and all regulations are evil for some reason, you know.
52:55
Yeah, I was in, it was a long
52:58
time ago, 2006, I was in Australia,
53:01
giving a talk for the Royal
53:02
Australasian College of Surgeons,
53:04
and part of the deal was that I would talk to the aviation
53:08
industry about simulation based training.
53:10
This was 2006, and one of the leaders
53:16
of the aviation simulation group come up,
53:18
he says to me, "Hold on a second." He
53:20
said, "You have all this validation data
53:23
about how effective it is in terms of skills training, and
53:27
you can't, it's not adopted?"
53:28
And I said, "No." And he said, "How come you've got it?" He
53:32
says, "We had no evidence when it was
53:34
mandated." Somebody in government
53:37
basically decided, "Okay, this is a better way."
53:41
And I mean, I don't know how we persuade them. A fellow of
53:46
mine said to me, "Tony, this isn't
53:49
data and evidence thing. This is a matter
53:52
of selling the idea to the powers that be."
53:58
So you do a lot of work and publish on
54:00
the developments in robotics and simulation.
54:04
Where do you envision that
54:05
simulation based training and robotics
54:08
are heading in the next 10 years or so?
54:11
Well, just like everything else, it's going to all change
54:14
completely because of AI, quite honestly.
54:18
I do think you'll see a lot more enhancements of things,
54:22
like the simulators because of AI
54:24
as time moves on. It's definitely going to happen. I think
54:30
it'll become more pervasive too,
54:31
because the threshold to do it... So I could go out and
54:37
build a new simulator now by myself
54:40
because of AI. I'm not going to do it because it doesn't
54:44
really make a lot of sense economically,
54:46
but I'd have the ability to do it because of how well Claude
54:50
and OpenAI Codex and all these other
54:53
things work now. It's going to dramatically drop the bar.
54:57
Robots are a different issue because you
54:59
still got to go through all the development of the
55:01
mechanics and things. It's not like AI is going
55:05
to make more surgical robots pop out quicker, but you're
55:09
going to see a lot more software enablement
55:11
in surgical robotics. So there'll be
55:14
little automations, little part-task automations
55:18
that are going to start showing up. Same in surgical
55:21
simulation, you'll see a lot more
55:24
sort of semi-automated instruction things are going to
55:27
start showing up independent of the
55:28
simulator cores themselves. Yeah, you're
55:31
just going to see a lot more of this for sure.
55:33
And how fast, do you think that autonomous robotics
55:39
performance surgical procedures will develop?
55:42
No, no, just no. So you've heard me say
55:49
this, other people have heard me say this,
55:50
I'm a real downer on this relative to academics because
55:55
they're all wanting to do this, is that
55:57
unless the financial structures are changed, there's no way
56:01
that autonomous surgery is going
56:03
to happen with commercial surgical robots because they're
56:07
not going to allow themselves to become
56:08
financially liable for what their robot does. As it is
56:13
right now, they just build the tool
56:15
and others use the tool and those other people are the ones
56:18
that become liable from use of that tool.
56:22
If they start automating things, then they're doing
56:26
surgery, they're performing medicine,
56:28
and that's a whole different world. You'll see lots of AI
56:32
that is going to be suggestive
56:34
of like, maybe you shouldn't do that, or maybe you should
56:37
notice this other thing there,
56:40
stuff like that. You'll see lots of that, but close the
56:44
wound for me. I don't think so.
56:49
You might have smart things. Your stapler example is a
56:52
really good one. There is a possibility that
56:55
some kind of a supervised stapler that just performs that
57:00
little tiny task and does it much
57:01
more uniformly and precisely than the human does. You're
57:07
not going to see autonomous procedures.
57:09
You could in a crazy situation like Satava always used to
57:14
talk about, like battlefield medicine,
57:16
or you're on your ship to Mars or whatever it is, then
57:20
totally, I get that. But in a general sense,
57:23
well, so there might be things in the developing world that
57:29
this might be a possibility to,
57:31
but I don't see it in the Western
57:35
Europe. All the developed markets, China, Asia,
57:43
Europe, the United States, I don't see it. I just don't.
57:47
Yeah. I'm sure that the clinicians will
57:50
have something to say about it as well.
57:52
I mean, I assume they're particularly aware of the
57:58
perceived threat of autonomous
58:01
operators or do they have a realistic understanding? This
58:05
is probably not going to happen.
58:07
Yeah. Like I said, I think it's a regulatory and a
58:12
financial issue and it's not a technology
58:15
issue. I think they will figure out how to
58:17
do it, but it's just like with self-driving,
58:22
right? The self-driving cars, they get freaked out by
58:25
things they haven't seen before, right?
58:28
And they've been trying really hard for a
58:30
really long time to get it to all be perfect,
58:33
and it's still not perfect. And you'd have to do the same
58:37
thing in surgery, and you just don't have
58:39
the infrastructure that you have for autonomous driving in
58:44
the first place, right? In terms of
58:45
data collection. And as we know, there's millions of miles
58:49
of simulated driving that's used to train
58:51
the autonomous driving systems. You'd have to do the same
58:55
thing for surgical robotics. And that
58:59
would mean that you'd have to verify the simulators being
59:02
accurate so that you know that the footage
59:05
you're synthetically generating is
59:07
actually reflective of what happens in real life.
59:09
So that might be the only way that we end up with good
59:12
surgical simulators. If somebody decides
59:14
they're going to invest a huge amount of money into
59:17
generating millions of hours of surgical video
59:22
through simulation, they'd have to actually verify the
59:25
Dickens out of it. So it might be the only way
59:27
you're ever going to get a properly verified surgical
59:29
simulator. But okay. So you would have
59:32
a map and the physical forces that are generated by the
59:37
different behaviors. I'm saying the
59:40
endovascular again, because I mean, you're working in a
59:44
confined space. And I suppose the
59:47
metrics for that there would be easier. I'm not saying
59:50
easy, but easier to generate.
59:53
But you would still need the metrics
59:56
of the performance of the procedure.
59:58
You know what the cardiologist or the radiologist does what they need to.
1:00:02
Yeah, so you'd be we'd be verified. We'd be building
1:00:06
physical stand ins for the for the
1:00:10
for the patient and then instrumenting the
1:00:12
hell out of them to get like the forces and
1:00:15
torques and pressures and everything else. And then we'd be
1:00:18
using those to verify the actual,
1:00:21
you know, virtual simulations, the other
1:00:24
computational simulations. And from there,
1:00:26
you would take those simulators used by clinicians, and
1:00:30
you'd probably be back mapping the you'd
1:00:33
have them perform the procedures people that everybody
1:00:35
agrees are good. And then you'd be
1:00:38
back mapping the information you pull out of those
1:00:41
simulators and turning those into metrics,
1:00:44
basically. So you don't think that the clinicians have
1:00:49
anything to worry about about an autonomous
1:00:51
robot anytime soon? No, I really
1:00:55
don't. I don't see it. Yeah. Okay.
1:00:59
You will see part task automation, but autonomous robots, no
1:01:03
Yeah, yeah. I mean, I think that's entirely feasible.
1:01:08
OK, Dwight, thanks for joining me and
1:01:11
Yeah Sure!
1:01:12
thanks for a very interesting conversation.
1:01:15
Yep, absolutely.