September 11, 2025 • 39 mins
Send us a text with a question or thought on this episode ( We cannot replay from this link)
Dr. Canio Martinelli, OBGYN specialist and head of clinical programs at Sbarro Health Research Organization, discusses groundbreaking research on AI applications in medicine and surgical decision-making to improve patient outcomes.
• Research shows AI systems like ChatGPT perform comparably to resident physicians in diagnostic accuracy
• Human doctors and AI make different types of errors, suggesting they could complement each other
• AI maintains consistent performance under time pressure while human performance declines
• The "gray area dilemma" in surgery refers to critical decisions surgeons make based on incomplete information
• PULSAR study aims to decode surgical decision-making by analyzing billions of data points
• True personalized medicine must consider what "functionality" means to each individual patient
• Future AI systems could help surgeons tailor procedures to each patient's specific anatomy and goals
• Communication remains challenging when explaining complex medical concepts and statistics to patients
Use promo code ENDOBATTERY for an exclusive 20% discount at Strong Coffee Company and help support these expert conversations at EndoBattery.
Website endobattery.com
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Alanna (00:01):
Welcome to Endo Battery Fast Charged, a series dedicated
to keeping you informed andempowered in the realm of
endometriosis.
Teaming up with board-certifiedpatient advocates, we bring you
the latest articles, researchand insights to equip you with
accurate information and adeeper understanding.
Whether you're expanding yourknowledge, staying updated or
seeking clarity, you're in theright place.
(00:21):
I'm your host, alana, and thisis Endo Battery Fast Charged
charging and empowering yourlife with knowledge.
Welcome back to Endo BatteryFast Charged, where we power
through the latest researchshaping endometriosis and
women's health.
I couldn't be more excited tohave our very first guest on
(00:46):
this series, Dr.
Canio Martinelli, an OBGYNspecialist and the head of
clinical program at SbarroHealth Research Organization at
Temple University.
Dr Martinelli, recently namedFDA AACR Oncology Educational
Fellow, is at the forefront oftranslating cutting-edge science
into real-world impact.
(01:06):
His work connects emergingresearch, clinical care and the
future treatment for people withendometriosis, helping us
better understand whereinnovation can truly change
lives.
And just as a friendly reminder, correlation does not equal
causation.
So let's keep our curiosityfully charged, but stay grounded
as we dig in.
(01:27):
You know how I always say grab acup of coffee or a cup of tea
and join me at the table beforewe get started.
Well, what if that cup could domore than just taste good?
Meet Strong Coffee Company, mylittle secret weapon.
It's premium instant coffeeloaded with protein, mcts and
adaptogens so you get smooth,energy, sharp focus and none of
(01:49):
that jittery crash.
Basically, it's coffee thatactually shows up for you.
Use my promo code endobatteryfor an exclusive 20% discount
and yes, when you do, you'rehelping me keep these expert
conversations brewing here atIndoBattery.
So go on, grab your cup, powerup by going to
strongcoffeecompanycom and usingthe code Endo Battery, and
(02:13):
let's make this episode evenbetter.
Let's get into this.
Thank you so much for joiningme and sitting down and going
over this research.
Let's start with how we doresearch.
Canio (02:24):
First of all, thank you for having me here.
You already know how much Iappreciate you.
I mean, whatever you do isincredible for patients, for
people and for us also,scientists and physicians,
because research, as we alreadysaid, is about trying to find
the right answer to people.
So we need to resonate together, to talk together, to be much
(02:47):
more connected when it comes toresearch.
We're talking about one of mygreatest passion, because
research is a way to findsolution to problems.
So every research starts withwhat we call research question.
So what is the question we aretrying to address?
And after we get that question,then we start to develop
(03:07):
different hypotheses and thenthe goal here is trying to have
the right methodology to confirmor not our hypothesis.
So it's everything extremelyserious and rigid, because at
the end, the conclusion that weget from that, out of this, has
to be applicable to people in aspecific way.
(03:30):
It means that we need to notjust give an information, just
give this study showed this.
We need to be able also to tellthem these are the data, this
is the way you have to applythat.
So it's then after we'vepublished the study.
That's what we call thecritical appraisal.
So how can I really use thatdata?
(03:51):
Is it going to use the data inthe same way in the world?
Is it applied only to specificpopulation, especially right now
that we are going inpersonalized and precision
medicine?
We will talk about that, but soI think that it's much easier
to talk about specific articles.
I can tell you all the process.
(04:11):
One of the most interesting forpeople study that we published
was the one on AI applied toimprove the clinical metrics for
physician.
Because everybody had thisexperience, like me as a
physician.
Also, explore CHATGPT toexplore rather than Claude,
Gemini, and, since access tohealthcare is very, very hard
(04:35):
today, as well as education formedical students, everybody,
somehow we relate with this newtechnology, how we relate with
this new technology.
So our idea is let's do thisscientific study in order to try
to understand not just whetherthis new machine, these new
tools, are better than physician, but I want to try to
(04:56):
understand if the reasoning ofthis machine is actually
different than physician,because if I understand that, I
know how to find a way to matchthem, to merge them, how to
really improve the ability forphysician.
But if I'm even able tounderstand how, the reason, I
can give practical tips topeople and say, okay, if you
(05:16):
really want to have a suggestion, a tip from Chachi PT or
Clodagh Gemini or Meta, you canask it, but you need to be aware
and to be able how to handlethat information.
That's the key, because as yougo to different doctors, they
can give you differentinformation.
Which one is the right one?
You can pick up the right onein front of a doctor because he
(05:37):
or she is nice or because youcan resonate with it when you do
with what we call largelanguage models.
So it's like very disconnectedstuff.
You don't have any, not justyou as a patient or people, but
you even as a doctor.
Right now.
You don't have any tips, anyguidelines to say, ok, what he's
saying is 100% sure, becauseyou don't know where the data
(05:58):
comes from.
So what we did?
We actually use different LLMsdifferent like ChatGPT, clove,
gemini, Meta, the differentversion and we choose 24
residents in a gynecology and wedevelop 60 different clinical
scenarios.
(06:19):
So we ask to the resident andwe ask to LLMs to try to
approach the clinical scenario,to give us an answer.
We did this in a differentsituation because the idea is
trying to see how can I use thattechnology and even to test the
technology in differentsituations.
So we decided to split thequestion following language.
(06:41):
So we test the scenarios inEnglish and in Italian, and then
we test the scenarios even withtime constraints.
It means that for some question, people were completely free to
answer, with all the calm ofthe world, the question.
In other scenarios they neededto give the same question, the
same answer in 10 seconds.
And then the resident, thegynecological resident.
(07:03):
They were actually grouped indifferent years of expertise, so
they were the fresh one and thesenior one.
That was a study that wasconducted last year, so during
2024,.
That's important to see becausethe technology, especially in
AI, is speeding up so much.
So maybe right now the durationcould be different.
(07:23):
But I actually did another testwhere I saw that the metrics
are pretty much the same.
So let's start from the firstoutcome, so the first results
from the study.
If we took overall, the all LLMversus the all resident across
all the different ages ofexpertise, we are basically
(07:45):
seeing that they match as aresult.
So right now, specifically tothat, 60 scenarios, ai is able
to give you almost the sameanswer of residents, not senior
physicians.
But then we did what we callsub-analysis, so within the data
we try to explore correlationin order to get practical
(08:08):
information.
So we understood, for example,that if you compare the senior
physician at a resident ingynecology to the average of
LLMs, the senior are much, muchbetter than the average LLM.
But if you take the best LLM,there was one that you need to
pay.
That one is almost sameaccuracy of the senior residents
(08:33):
.
But the other sub-analysis wasduring time constraints, because
when you put pressure on peopleand residents you see that
their performance metrics gonnareally reduce a lot, while the
LLM metrics are staying stable.
And then we also saw thepattern, what we call the
pattern of mistake of humanscompared to machines, and we saw
(08:54):
that the pattern of mistakes isactually different.
So that's important becausewhat you do as a physician or a
people like, whenever youinquire those machines, they
come to a conclusion thedifferent one you take.
So who is the one that is right, me or the machine?
Alanna (09:10):
Right.
Canio (09:11):
The problem is that being able to predict whether a
machine is wrong is as much ashard than being able to predict
whether a physician is wrong.
So the real take from the studyis maybe for people right now
are able to give you generaladvices in the right way, but if
(09:32):
you really want to have theanswer from yourself, it's still
not well performed.
Because if you want to knowsomething about anything issue,
any field, you want to have akind of recommendation, it could
be okay using the stuff.
But if you have a problem andyou want to have a solution that
is really perfect for yourselfin your specific time, your
(09:53):
specific situation, well that'snot the right tool.
And when it comes for physician, it depends the way you use LLM
, because basically this stuffis just an advanced researcher.
It depends the way you use LLM,because you basically this
stuff is just an advancedresearcher.
It depends the way you writethe prompt.
So in the way you give them thedata to analyze the case, the
more option the machine has,it's easier for them, for the
(10:28):
machine to make a mistake.
And that's where it comes.
Expert physician, expertphysician, the one that's not
just you know, mastering onetechnique and offering only one
technique to patient or onetreatment.
The great physician is one thatknows a lot of different kind
of management and is able toapply that specific management
(10:50):
or treatment to your specificcondition of life.
So in this matching course, youknow, senior physicians are so
much better than LLM.
But here again, we don't needLL to substitute humans.
We need LLM in maybe, forexample, when you have pressure,
when you have a high workload,you need to take a decision.
(11:13):
It can help you to give thedata that you need to take the
decision.
So it's much more like aco-piloting and when it comes to
people, it can be an assistantto check on it, like, okay, did
you check this, did you?
But not just for giving medicaladvices, but in order to get
the data that then will beincorporated by the system and
(11:34):
the physician is able to trackthe data in a more clear way.
Here is all about having thechance to see the pattern.
The pattern of the data comingfrom a patient is critical for
the physician to act in the bestway.
So this first study basically,is showing that if we need to
(11:55):
train a system, we know that theway AI and doctors so humans
reason is different.
We need to be aware of thesedifferences and then, when we
want to implement a system thatis AI powered with human in the
loop, it means that there is asystem that is taking care about
you you need AI that is calledsupervised.
(12:18):
So it means that I need toteach the AI whatever I know and
I need to set boundaries,because these LLM in the studies
were like commercial LLM, soeverybody can get access to that
.
The next step is going to bebeing able to use that
technology but, more importantly, being able to teach and set
(12:39):
boundaries.
So it says you cannot go toofar away from that.
If the patient really want tohave that kind of answer, you're
not able to give it.
So just says you cannot go toofar away from that.
If the patient really want tohave that kind of answer, you're
not able to give it.
So just say refer to thedoctors and we need them to
train time after time, yearafter year and in a way that it
can take care of the mostannoying stuff and humans can be
much more oriented.
(13:00):
They can be used in verycomplex scenarios, but overall
that's even a challenge forhumans, actually, because you
are understanding here that as ahuman, you need to go faster
here.
As a physician, you need to beextremely trained to think.
You need to be extremelytrained not to analyze the data,
(13:20):
but to talk with people,because they are getting to us.
I'm not saying that I will makea diagnosis.
I'm saying that if we do notevolve as humans, ai will get
better than us.
That's for sure, because it'sall around the data and this
study is showing that Average AIis basically average resident
in a gynecology.
(13:40):
It's kind of you know, pusheven for next generation, young
generation.
You need to do better than mebecause I'm going to build a
system that better than me.
If you're not better than me,my system is going to be better
than you as a physician, and soit's kind of a research.
It's kind of scary because ifyou look at the data, basically
what we're showing is thatactual LMM can be even useful in
(14:05):
learning for physicians.
But I want to give hope.
You know that if you read thediscussion of the paper, it's
all about how can I get thisdata to tailor the new AI model
to help physicians.
And with those data now we'regoing to the next step.
That is the ongoing paper thatwe are basically writing right
(14:25):
now.
I can't say the name of thestudy because actually it's
going to be published very soon.
It's called Pulsar Study.
It's a study where we are rightnow building the architecture,
the digital human architecture,of the system that we will
develop in the future.
And that's very cool because itcomes from my heart.
(14:48):
My passion is I don't know ifyou have any question for the
other study, otherwise I jump inthis new.
Alanna (14:56):
Well, I think I want something to point out in this
study and maybe you touch onthis more.
This isn't necessarily areplacement for doctors.
It's a tool to be helpful forproviders and for patients, not
a replacement for those thatmight be a little bit scary.
You're saying that this studyis emphasizing that it's not
taking over, it's making thembetter.
Canio (15:16):
Yes, absolutely.
You know, I know educationmedical in healthcare is crazy,
but that's a cool thing.
We need to keep pushing,because whenever we understand,
whenever actually we explorethis system, we are basically
exploring a way to understandhow people reason even more,
(15:39):
because whatever we are doing inthe computer science applied to
healthcare is actually justtranspose what humans already do
in their brain and build aartificial system that is able
to do basically the same stuffand the same way.
While we are exploring this newtechnology, what's happening is
(16:02):
that we are even exploringourselves much better.
So it's a way that is kind of achallenge, because the more you
develop it, the more you havenew connection in your mind, and
AI right now is not able tofind something that is not
possible, to find Everything newthat it finds is not
necessarily true.
Same things in the researchright, whatever we find, it's
(16:23):
not necessarily true.
As same things in the researchright, whatever we find, it's
not necessarily true.
We need them to validate and,in real people, to see it, to
show that it's really, reallygood.
So in doing this, ai isextremely brilliant.
When we tell them exactly avery limited system, we say
that's how we work, we give yourules and that's where you have
(16:45):
to act Unlikely.
Right now we're not talkingabout ChatGPD or anything
commercial available.
It's a system that uses theirtechnology, but they're a system
underdeveloped, so it's stillnot reality.
There are also lots of studiesthat are showing that most of
the AI tools that have beendeveloping in healthcare, in the
first stages phases they arebrilliant.
(17:07):
Then, when they apply the toolsto real situation, real
hospital, they really do notperform well.
Yes, that's because we need tounderstand, we need to validate
those stuff, so it's notnecessarily right.
So the science is cool becauseit's kind of remember that you
need to be humble, you know,still need to consider that
(17:29):
you've wrong.
Your research question may bewrong.
If it's wrong, it doesn't meanthat is bad.
It means that that is not theright way.
You need to find another one.
And from that perspective, weare getting to like by make
failure and mistake.
We are getting to like buyingmake failure and mistake.
We are getting to somethingmuch, much better and it's going
to happen.
It's happening very, very, veryfast.
Alanna (17:50):
Interesting.
So did this research thentranslate into what your next
paper that's going to bepublished Did it?
Was that the catalyst to thisnext paper, Exactly?
Canio (18:03):
Absolutely the way we, you know, whenever we do
research, let's say, even needsto resonate with the money and
the funding that we have.
So we need to be strategic inwhat we do because we need to
worth every penny, every dollarwe have.
So we need to move in thatdirection and every evidence
that whenever you publish apaper, you, whatever people read
(18:25):
, there are data.
But behind that paper there isa team of people that recently
talk, so they make briefing andmeeting, they analyze data,
contextualize data within theactual medical knowledge and the
future and the different otherscience.
How can I do that?
So it's everything youexperience coming from the paper
(18:45):
.
So that's why wheneverscientists publish paper,
they're excited.
Maybe nobody's going to read thepaper when it comes to people
like, not the medical community,but for us it's amazing because
behind that paper there areideas, confirmed, dismissed.
People get gut hungry, you know.
And then there is all the allof this stuff.
(19:06):
So there are humans, life,working for something and
challenging each other to dosomething different and change
the actual like management.
So it helped us to jump to theother program that actually is
not just this paper, but it isour history, other papers, like
they merged together and theyjumped in this new paper, that
(19:30):
is, a theoretical paper withsome computational predictions
or some preliminary data.
That is called PULSAR.
That's basically meaningprobabilistic and user-centered
learning for surgical adaptiverecommendation.
Alanna (19:46):
That's a long word for a lot of us, oh my gosh, alan.
Canio (19:50):
I think I could speak about that for hours, because
it's really what really drivesme.
So please stop me if I keepgoing like without any direction
.
But it's really something thatI, you know, whenever I go to
sleep.
Alanna (20:12):
I think about that, believe me.
So the research is like.
The words in the title itselfare more manpower than I want to
put in half the time just tothink of what it is.
What is this paper, though?
What does it entail?
Because I'm sure that it's likea new pathway to what the
future holds forpatient-centered care.
Canio (20:27):
Okay, Maybe we can talk about that by having a
conversation.
I want to try to introduce thethings by using metaphor.
Alanna (20:36):
Okay.
Canio (20:38):
So let's say that, do you like cooking, love it.
So whenever you cook, do youfollow a specific recipe, the
specific steps, or you go byyour mind?
A little bit of both.
Okay, exactly, exactly.
I love this answer.
So sometimes you change therecipe, you change ingredients.
The question here is how andwhat is the things that you
(20:59):
follow to change your mind aboutsome step.
You know what I mean.
Like, how do you?
We're talking about yourdecision-making process in your
mind.
So whenever you decide tochange something in your like,
let's say, management of thedish, what is the things that
(21:20):
really drive you?
Alanna (21:21):
If I like the flavor of something or if I don't have the
right ingredient, but somethingcan substitute it, so like
knowing what I have available tome versus what I want, if that
makes sense.
Canio (21:33):
Okay, it makes sense.
But you know, maybe you canopen your kitchen and you find a
hundred ingredients.
How do you pick especiallyspecifically that one instead of
another one?
Alanna (21:43):
What might work better in my mind, like what I feel
like would make it better.
Canio (21:49):
That's based on what your feeling is based on what?
Alanna (21:52):
Yeah?
Canio (21:52):
And what's your opinion is based on?
Alanna (21:54):
The way I feel or the way I prefer things.
Exactly yeah.
Canio (21:59):
Now we are entering what we call in this paper the gray
area dilemma.
So there is a gray area in ourmind of decision-making process,
especially in surgery, whereactually we do something.
Most of the process isevidence-based, but sometimes
during this process there arelittle choices in every surgery.
(22:20):
Every surgery is completelydifferent than anyone else and
from anyone else, because everypatient is completely different
than each other.
So then, surgery is a specificfield of medicine because it's
the same time diagnosis andtreatment, because you don't
have all the data that youreally need before the surgery.
Most of the time you discoversomething when you open the
(22:42):
patient and when you're thereyou're making diagnosis and you
have to change, sometimes, yourmind.
That's why, when we shareinformed consent, there is a
huge list of something that canhappen, like we are doing.
A surgeon that maybe you knowfor endometriosis they say I can
resect the bowel, you can comeup with a stoma, we can resect
ureter, it may be possible thatwe have a urinary stoma, you
(23:03):
know, but there is even a riskof fatality.
That's because we don't have aclear idea what's happening 100%
before opening the patient.
When you are there, you reallyhave the.
You know the reality, so youneed to adapt your
decision-making process as asurgeon between evidence coming
(23:24):
from science and then somethingthat you have in your mind
that's based on your experience,your intuition, something we're
still not able to identify.
That's the gray area dilemma,and the best surgeon is the one
that somehow has that gray areathat is much more efficient and
(23:44):
effective when, at the end, ittakes the choice for the patient
during surgery.
It doesn't mean that it'snecessarily the one that is the
oldest one, but it's the onethat the sum of his option is
actually the result as a result,as the best outcome for the
patient is actually the resultas a result, as the best outcome
for the patient.
(24:05):
And that's why surgery isextremely hard to teach, rather
than compared to any other fieldof medicine, because it's hard
to explain, objectify andstructure the gray area of the
decision making of the surgeon.
Now, where this study, you know,fit, this study fit in, uh in
in all of this.
So the thing here is I needsomeone, since, starting from
(24:26):
the previous study where wewanted to try to codify the uh
decision-making pattern of themachine, here is trying to
codify the decision-makingprocess of the human by using AI
.
The goal is using AI technologyUh, I'm not going to go into
the details in this but thehuman by using AI.
The goal is using AI technology.
I'm not going to go intodetails in this, but the goal is
using AI in order to detect andto study how the surgeon
(24:50):
behaves in all the differentscenarios, all the different
surgeries.
But the most important thing isthen to anchor and to attach to
this reasoning and evaluationwhat's happening to the patient,
because we need to anchor thealgorithm to the outcome for the
(25:11):
patient, because the matrixhere is I can judge whether the
action of a surgeon was betterthan the other one by relating
his action or her action to theoutcome of the patient.
And it's not just that, becausethen any patient is different.
Someone could say exactly thegoal is try to destroy them.
(25:33):
Let me say that the patient inall little pieces.
Those are data before thesurgery, with all the imaging,
objective evaluation, clinicalexamination, and so we get some
of the data.
Then, during the surgery, wehave a video, so we get
information coming from thevideo getting during the surgery
(25:54):
, and that's another package ofdata.
The other thing that's veryinteresting is that whenever you
do surgery, you are modifyingthe anatomy.
You are modifying the biologyof the body, so what's happening
is that we need to be able torelate our action to the
modification in that specificpatient.
Science is always working onunderstanding the differences,
so all of this little actionwill generate differences
(26:17):
between the patient, and so atthe end you have to like try to
picture that at one patientduring all the process, so
assessed in a dynamic way, sopre, during surgery, after
surgery, will be actuallyrepresented by billions of data.
And these data are actuallybelonging they're not like
random they are belonging tospecific compartments and that's
(26:40):
a called layer, and that'swhere actually multi-army comes
in.
So we don't just need to getthe data, but we need to
organize in their owncompartment and then we need to
be able to relate them in astructured way.
Each of them needs to be likepart of a compartment and then
we need to in this billion ofdata.
(27:02):
We need that to understand whatwas the strategy that actually,
in that specific case, helpedthe surgeon to get to the best
result for the patient.
And then in the next case, themachine would be.
After doing that, with differentphysicians and different people
, the machine would be able toguide us and also the patient,
(27:22):
by little by little, say okay,if you do this, be careful,
because last time you did that,you did the same stuff.
Although it's anatomically andtechnically correct, let me say
that was not good extremely goodfor the patient.
Because whatever we need to dois to anchor this to the patient
outcomes.
Most of the time if you talk tophysicians, to surgeons, they
(27:44):
say I did this and it wasperfect.
The other one did another stuffthat was indeed perfect.
So here is not about doing whatphysician or surgeon says is
perfect, but is to anchor any,every single step to the patient
outcomes.
And that is going to improveeven understanding of how a
surgeon thinks.
(28:05):
In that gray area is evenmedical education and also in
the shared decision making,because at this way, after some
times, before the patient comesto me, you know, before the
surgery I can, by looking at thedata, say okay, that's the best
choice for you among the allpossible ones and it's going to
give you that specific risk andthat specific benefit.
Alanna (28:28):
Interesting.
So it's like a, it's a guidefor the doctor to give the
patient the best outcome whilealso giving, like the patient, a
voice in the long-term outcome.
Exactly, exactly.
That's great and that's likesomething, because I think we
always say it's not aone-size-fits-all, especially
with endometriosis surgery, andso just tailoring the approach
(28:52):
to the actual patient prior tosurgery for better outcomes
long-term, is that kind of Wow,I was so confused when I
explained all of this and yougot everything very clearly.
Canio (29:05):
I don't know how you did it.
I was so.
It was a mess when I said that,because I follow my inner flow,
you know what I mean.
Alanna (29:12):
No, I picked it up.
I just am like fascinated bythe fact that you know, as
patients, we are constantlylooking at okay, like, is this
bowel resection right for me, oris this whatever resection
right for me, or is thiswhatever Is it right for me?
And this is just a good, moreeducated way of approaching a
surgery and making it trulymultidisciplinary from, like,
(29:33):
the AI standpoint, mixed withthe provider, mixed with the
patient, the outcome beingbetter.
This is crazy to me.
Canio (29:41):
What you said.
See, this conversation isbeautiful for this reason,
because it's I'm learningsomething from you and I or
treatment, we take our decisionby looking at data coming from.
We say, population study, soall the data we have.
(30:15):
We say, ok, you have to do thatbecause the chance of success
is this or because it's best foryou.
But OK, the question is how, asa patient, how am I similar to
the population that was using inthe study that generate the
evidence?
And that's where medicine isnot precise, because the
evidence are coming from a setof data that not necessarily are
(30:38):
the same of the very nextpatient that is in front of me.
And when it comes to surgery,it's even more difficult,
because I don't give the samepills to everybody, but actually
I'm tailoring the surgicalprocedure to the specific
patient and that's unique in anypatient.
So, whatever you introduce adecision and the differences,
you are making thedecision-making process much
(31:01):
more complicated.
Alanna (31:02):
Yeah.
Canio (31:03):
And very hard to understand.
Alanna (31:04):
Is this what?
When you did this paper, werethey seeing video or AI like
virtual reality ways of doingthe surgery, or was it just
words Like was it just feedback?
Canio (31:16):
through.
That's a very, very interestingquestion because it's like it's
actually a very detailedquestion and so I need to jump
necessarily to details.
So in order to design, first ofall, you start by designing the
algorithm that is moreefficient, Because in order to
(31:37):
do something, you can do it indifferent ways.
Then you have to find the onethat really is tuned in the
right way.
So this starts as a theoreticalstudy coming from observation
that I had personally in my lifeexperience as a physician and
scientist, Because one of thefantastic things you know, if
(32:00):
you really want to judge I don'twant to say judge, but if you
really want to have the realfeeling of a surgeon, I think
it's based on how really hetraveled in his life.
So because as much she has orshe has been exposed to
something different, then he canunderstand really what's the
best things.
(32:20):
Because whenever you are in thebest center in the world,
they're going to teach youthat's the right method and
that's okay because it comesfrom a school.
But you need to get to thatconclusion after you see
different methods and you startto believe actually, that is to
see that there are surgeons thatare doing different kind of
techniques and they still havetreating in the best way
(32:42):
possible and patients are happyabout that.
So the thing is coming from twoproblems.
So evidence, real problem,where you see all of these, a
surgeon in school says that theyare doing the best things for
the patient.
And then methodological problem.
That's coming from my researchfield, where actually you have
to question everything you do.
And how do you stop questioningwhat you do when you have no
(33:05):
more grade zones?
How do you not have grade zones?
You start to get the data andstart to, mathematically
speaking, give a system for eachof the data you come in game.
So the question is what kind ofdata are there?
The kind of data, every kind ofdata.
So they comes from imaging,they come from talking, so word
(33:26):
they come from.
It's called finite elementanalysis.
So what this conceptualizationthinks about is that we get the
data from ultrasound or RMI andwe translate those data that are
basically pixel in a system, acomputer that is able to
(33:46):
simulate the real functioning ofthe anatomy, but in a
functional way, so you cansimulate what's happening.
You say, okay, I want to dothis procedure.
You simulate the procedure andyou see that something else is
happening and that's why you cantailor the surgery before doing
the surgery, and then it'sgoing to tell you that the
(34:07):
things you're doing is good.
Well, this is easy when it comesto cancer.
So oncological surgery, becausethe goal is overall survival.
But when it comes to function,it's a completely different
stuff, Because how do you reallyassess functionality in women
today?
By some questionnaire what isfunctionality?
It's actually really.
It's different for every woman.
(34:28):
So, for example, I can say thatfor a young athlete woman,
functionality is being able tohave no pain and to be as much
as she can and give the bestperformance.
For an old lady, maybe thefunctionality is able to hold
(34:50):
the hands of her grandson.
So even if she's pained, it'snot a problem.
So is everything based on thisquestion?
How can I representmathematically people's problem
and human's problem?
Because whatever you aresearching, you're human acting
and not human.
And so when the study will bepublished, the cool things are
(35:14):
going to be the architecture ofall of this, because it's going
to be extremely detailed andevery step is connected to
another one.
Alanna (35:22):
And then there are loops that reinforce and retune all
the system again connect it toanother one, and then there are
loops that reinforce and retuneall the system again.
It's like taking both thevisual of the scans and then
you're taking the patient.
Does it account for, likepatient symptoms?
Does it account for, like allof the functional things that
are at that point failing themfor lack of a better word and
(35:42):
taking that into considerationand giving you or the provider
the approach that's going tobest suit them long term in a
visual manner?
Or is it just like spelling itout, it's like a checklist, or
is it like a visual aid for them?
Canio (35:59):
And that's where it comes .
Another project, right, andthat makes me feel that we are
going in the right direction.
This is something very, very,you know, exclusive, preliminary
, because here comes the secondpoint communication.
And do you really understandwhether what you're telling
(36:22):
people patient you're reallyable to explain in a clear way?
Are you able, really you know,whenever you say the chance of
success is 80% and the failure20%?
Okay, and the patient said thatevery patient will believe
they're going to be in the 20%.
So there is a problem here, evenin communication, and that's
(36:43):
another part of the problem thatis not focused in this paper,
but it's another project we areleading, that's translational
communication in the medicine,and we are working together with
visual artists, so they arehelping us in trying to
understand how to translatecomplicated medical stuff in
(37:04):
something that people can reallyunderstand, but not logically
understand, even emotionallyunderstand.
They need to understand that,because when you say 80, 20% are
very, you know, great data, butthe 80% is not just the 80%,
it's also what's happening ifyou don't do that kind of
(37:25):
procedure.
So if you don't do that kind ofprocedure, so if you don't
choose that procedure, somethingelse is going to happen.
So it's giving the right let'ssay even emotional way to pure
great data?
The answer for me I don't know.
I'm working together with thevisual artists, so people that
work with that are inspired byemotion much more than me, that
(37:50):
have been exposed in life tosomething different than me.
So in science and that's thebeauty of the United States, I
have to tell this Whenever youwork in university, you have all
of these academic peopleworking together and give their
point of view and really, at theend, try to do something that
is really effective, not justbeautiful chrome.
It needs to work out.
Alanna (38:10):
Right, I'm excited to see where that one goes, because
I feel like that could changethe narrative of outcomes.
Canio (38:17):
Yes, which?
Alanna (38:17):
is exciting.
That's a wrap for EndoBattery'sFast Charge this week.
I hope this episode left youinspired and empowered to
continue advocating for yourcare and encourage that research
is happening and change canhappen in our lifetime.
Make sure to join us next timeas Kanyo sits down with us to
explore more research that'sbeing done.
(38:39):
Until then, continue advocatingfor you and for others.
Welcome to Endo Battery Fast Charged, a series dedicated
to keeping you informed andempowered in the realm of
endometriosis.
Teaming up with board-certifiedpatient advocates, we bring you
the latest articles, researchand insights to equip you with
accurate information and adeeper understanding.
Whether you're expanding yourknowledge, staying updated or
seeking clarity, you're in theright place.
(00:21):
I'm your host, alana, and thisis Endo Battery Fast Charged
charging and empowering yourlife with knowledge.
Welcome back to Endo BatteryFast Charged, where we power
through the latest researchshaping endometriosis and
women's health.
I couldn't be more excited tohave our very first guest on
(00:46):
this series, Dr.
Canio Martinelli, an OBGYNspecialist and the head of
clinical program at SbarroHealth Research Organization at
Temple University.
Dr Martinelli, recently namedFDA AACR Oncology Educational
Fellow, is at the forefront oftranslating cutting-edge science
into real-world impact.
(01:06):
His work connects emergingresearch, clinical care and the
future treatment for people withendometriosis, helping us
better understand whereinnovation can truly change
lives.
And just as a friendly reminder, correlation does not equal
causation.
So let's keep our curiosityfully charged, but stay grounded
as we dig in.
(01:27):
You know how I always say grab acup of coffee or a cup of tea
and join me at the table beforewe get started.
Well, what if that cup could domore than just taste good?
Meet Strong Coffee Company, mylittle secret weapon.
It's premium instant coffeeloaded with protein, mcts and
adaptogens so you get smooth,energy, sharp focus and none of
(01:49):
that jittery crash.
Basically, it's coffee thatactually shows up for you.
Use my promo code endobatteryfor an exclusive 20% discount
and yes, when you do, you'rehelping me keep these expert
conversations brewing here atIndoBattery.
So go on, grab your cup, powerup by going to
strongcoffeecompanycom and usingthe code Endo Battery, and
(02:13):
let's make this episode evenbetter.
Let's get into this.
Thank you so much for joiningme and sitting down and going
over this research.
Let's start with how we doresearch.
Canio (02:24):
First of all, thank you for having me here.
You already know how much Iappreciate you.
I mean, whatever you do isincredible for patients, for
people and for us also,scientists and physicians,
because research, as we alreadysaid, is about trying to find
the right answer to people.
So we need to resonate together, to talk together, to be much
(02:47):
more connected when it comes toresearch.
We're talking about one of mygreatest passion, because
research is a way to findsolution to problems.
So every research starts withwhat we call research question.
So what is the question we aretrying to address?
And after we get that question,then we start to develop
(03:07):
different hypotheses and thenthe goal here is trying to have
the right methodology to confirmor not our hypothesis.
So it's everything extremelyserious and rigid, because at
the end, the conclusion that weget from that, out of this, has
to be applicable to people in aspecific way.
(03:30):
It means that we need to notjust give an information, just
give this study showed this.
We need to be able also to tellthem these are the data, this
is the way you have to applythat.
So it's then after we'vepublished the study.
That's what we call thecritical appraisal.
So how can I really use thatdata?
(03:51):
Is it going to use the data inthe same way in the world?
Is it applied only to specificpopulation, especially right now
that we are going inpersonalized and precision
medicine?
We will talk about that, but soI think that it's much easier
to talk about specific articles.
I can tell you all the process.
(04:11):
One of the most interesting forpeople study that we published
was the one on AI applied toimprove the clinical metrics for
physician.
Because everybody had thisexperience, like me as a
physician.
Also, explore CHATGPT toexplore rather than Claude,
Gemini, and, since access tohealthcare is very, very hard
(04:35):
today, as well as education formedical students, everybody,
somehow we relate with this newtechnology, how we relate with
this new technology.
So our idea is let's do thisscientific study in order to try
to understand not just whetherthis new machine, these new
tools, are better than physician, but I want to try to
(04:56):
understand if the reasoning ofthis machine is actually
different than physician,because if I understand that, I
know how to find a way to matchthem, to merge them, how to
really improve the ability forphysician.
But if I'm even able tounderstand how, the reason, I
can give practical tips topeople and say, okay, if you
(05:16):
really want to have a suggestion, a tip from Chachi PT or
Clodagh Gemini or Meta, you canask it, but you need to be aware
and to be able how to handlethat information.
That's the key, because as yougo to different doctors, they
can give you differentinformation.
Which one is the right one?
You can pick up the right onein front of a doctor because he
(05:37):
or she is nice or because youcan resonate with it when you do
with what we call largelanguage models.
So it's like very disconnectedstuff.
You don't have any, not justyou as a patient or people, but
you even as a doctor.
Right now.
You don't have any tips, anyguidelines to say, ok, what he's
saying is 100% sure, becauseyou don't know where the data
(05:58):
comes from.
So what we did?
We actually use different LLMsdifferent like ChatGPT, clove,
gemini, Meta, the differentversion and we choose 24
residents in a gynecology and wedevelop 60 different clinical
scenarios.
(06:19):
So we ask to the resident andwe ask to LLMs to try to
approach the clinical scenario,to give us an answer.
We did this in a differentsituation because the idea is
trying to see how can I use thattechnology and even to test the
technology in differentsituations.
So we decided to split thequestion following language.
(06:41):
So we test the scenarios inEnglish and in Italian, and then
we test the scenarios even withtime constraints.
It means that for some question, people were completely free to
answer, with all the calm ofthe world, the question.
In other scenarios they neededto give the same question, the
same answer in 10 seconds.
And then the resident, thegynecological resident.
(07:03):
They were actually grouped indifferent years of expertise, so
they were the fresh one and thesenior one.
That was a study that wasconducted last year, so during
2024,.
That's important to see becausethe technology, especially in
AI, is speeding up so much.
So maybe right now the durationcould be different.
(07:23):
But I actually did another testwhere I saw that the metrics
are pretty much the same.
So let's start from the firstoutcome, so the first results
from the study.
If we took overall, the all LLMversus the all resident across
all the different ages ofexpertise, we are basically
(07:45):
seeing that they match as aresult.
So right now, specifically tothat, 60 scenarios, ai is able
to give you almost the sameanswer of residents, not senior
physicians.
But then we did what we callsub-analysis, so within the data
we try to explore correlationin order to get practical
(08:08):
information.
So we understood, for example,that if you compare the senior
physician at a resident ingynecology to the average of
LLMs, the senior are much, muchbetter than the average LLM.
But if you take the best LLM,there was one that you need to
pay.
That one is almost sameaccuracy of the senior residents
(08:33):
.
But the other sub-analysis wasduring time constraints, because
when you put pressure on peopleand residents you see that
their performance metrics gonnareally reduce a lot, while the
LLM metrics are staying stable.
And then we also saw thepattern, what we call the
pattern of mistake of humanscompared to machines, and we saw
(08:54):
that the pattern of mistakes isactually different.
So that's important becausewhat you do as a physician or a
people like, whenever youinquire those machines, they
come to a conclusion thedifferent one you take.
So who is the one that is right, me or the machine?
Alanna (09:10):
Right.
Canio (09:11):
The problem is that being able to predict whether a
machine is wrong is as much ashard than being able to predict
whether a physician is wrong.
So the real take from the studyis maybe for people right now
are able to give you generaladvices in the right way, but if
(09:32):
you really want to have theanswer from yourself, it's still
not well performed.
Because if you want to knowsomething about anything issue,
any field, you want to have akind of recommendation, it could
be okay using the stuff.
But if you have a problem andyou want to have a solution that
is really perfect for yourselfin your specific time, your
(09:53):
specific situation, well that'snot the right tool.
And when it comes for physician, it depends the way you use LLM
, because basically this stuffis just an advanced researcher.
It depends the way you use LLM,because you basically this
stuff is just an advancedresearcher.
It depends the way you writethe prompt.
So in the way you give them thedata to analyze the case, the
more option the machine has,it's easier for them, for the
(10:28):
machine to make a mistake.
And that's where it comes.
Expert physician, expertphysician, the one that's not
just you know, mastering onetechnique and offering only one
technique to patient or onetreatment.
The great physician is one thatknows a lot of different kind
of management and is able toapply that specific management
(10:50):
or treatment to your specificcondition of life.
So in this matching course, youknow, senior physicians are so
much better than LLM.
But here again, we don't needLL to substitute humans.
We need LLM in maybe, forexample, when you have pressure,
when you have a high workload,you need to take a decision.
(11:13):
It can help you to give thedata that you need to take the
decision.
So it's much more like aco-piloting and when it comes to
people, it can be an assistantto check on it, like, okay, did
you check this, did you?
But not just for giving medicaladvices, but in order to get
the data that then will beincorporated by the system and
(11:34):
the physician is able to trackthe data in a more clear way.
Here is all about having thechance to see the pattern.
The pattern of the data comingfrom a patient is critical for
the physician to act in the bestway.
So this first study basically,is showing that if we need to
(11:55):
train a system, we know that theway AI and doctors so humans
reason is different.
We need to be aware of thesedifferences and then, when we
want to implement a system thatis AI powered with human in the
loop, it means that there is asystem that is taking care about
you you need AI that is calledsupervised.
(12:18):
So it means that I need toteach the AI whatever I know and
I need to set boundaries,because these LLM in the studies
were like commercial LLM, soeverybody can get access to that
.
The next step is going to bebeing able to use that
technology but, more importantly, being able to teach and set
(12:39):
boundaries.
So it says you cannot go toofar away from that.
If the patient really want tohave that kind of answer, you're
not able to give it.
So just says you cannot go toofar away from that.
If the patient really want tohave that kind of answer, you're
not able to give it.
So just say refer to thedoctors and we need them to
train time after time, yearafter year and in a way that it
can take care of the mostannoying stuff and humans can be
much more oriented.
(13:00):
They can be used in verycomplex scenarios, but overall
that's even a challenge forhumans, actually, because you
are understanding here that as ahuman, you need to go faster
here.
As a physician, you need to beextremely trained to think.
You need to be extremelytrained not to analyze the data,
(13:20):
but to talk with people,because they are getting to us.
I'm not saying that I will makea diagnosis.
I'm saying that if we do notevolve as humans, ai will get
better than us.
That's for sure, because it'sall around the data and this
study is showing that Average AIis basically average resident
in a gynecology.
(13:40):
It's kind of you know, pusheven for next generation, young
generation.
You need to do better than mebecause I'm going to build a
system that better than me.
If you're not better than me,my system is going to be better
than you as a physician, and soit's kind of a research.
It's kind of scary because ifyou look at the data, basically
what we're showing is thatactual LMM can be even useful in
(14:05):
learning for physicians.
But I want to give hope.
You know that if you read thediscussion of the paper, it's
all about how can I get thisdata to tailor the new AI model
to help physicians.
And with those data now we'regoing to the next step.
That is the ongoing paper thatwe are basically writing right
(14:25):
now.
I can't say the name of thestudy because actually it's
going to be published very soon.
It's called Pulsar Study.
It's a study where we are rightnow building the architecture,
the digital human architecture,of the system that we will
develop in the future.
And that's very cool because itcomes from my heart.
(14:48):
My passion is I don't know ifyou have any question for the
other study, otherwise I jump inthis new.
Alanna (14:56):
Well, I think I want something to point out in this
study and maybe you touch onthis more.
This isn't necessarily areplacement for doctors.
It's a tool to be helpful forproviders and for patients, not
a replacement for those thatmight be a little bit scary.
You're saying that this studyis emphasizing that it's not
taking over, it's making thembetter.
Canio (15:16):
Yes, absolutely.
You know, I know educationmedical in healthcare is crazy,
but that's a cool thing.
We need to keep pushing,because whenever we understand,
whenever actually we explorethis system, we are basically
exploring a way to understandhow people reason even more,
(15:39):
because whatever we are doing inthe computer science applied to
healthcare is actually justtranspose what humans already do
in their brain and build aartificial system that is able
to do basically the same stuffand the same way.
While we are exploring this newtechnology, what's happening is
(16:02):
that we are even exploringourselves much better.
So it's a way that is kind of achallenge, because the more you
develop it, the more you havenew connection in your mind, and
AI right now is not able tofind something that is not
possible, to find Everything newthat it finds is not
necessarily true.
Same things in the researchright, whatever we find, it's
(16:23):
not necessarily true.
As same things in the researchright, whatever we find, it's
not necessarily true.
We need them to validate and,in real people, to see it, to
show that it's really, reallygood.
So in doing this, ai isextremely brilliant.
When we tell them exactly avery limited system, we say
that's how we work, we give yourules and that's where you have
(16:45):
to act Unlikely.
Right now we're not talkingabout ChatGPD or anything
commercial available.
It's a system that uses theirtechnology, but they're a system
underdeveloped, so it's stillnot reality.
There are also lots of studiesthat are showing that most of
the AI tools that have beendeveloping in healthcare, in the
first stages phases they arebrilliant.
(17:07):
Then, when they apply the toolsto real situation, real
hospital, they really do notperform well.
Yes, that's because we need tounderstand, we need to validate
those stuff, so it's notnecessarily right.
So the science is cool becauseit's kind of remember that you
need to be humble, you know,still need to consider that
(17:29):
you've wrong.
Your research question may bewrong.
If it's wrong, it doesn't meanthat is bad.
It means that that is not theright way.
You need to find another one.
And from that perspective, weare getting to like by make
failure and mistake.
We are getting to like buyingmake failure and mistake.
We are getting to somethingmuch, much better and it's going
to happen.
It's happening very, very, veryfast.
Alanna (17:50):
Interesting.
So did this research thentranslate into what your next
paper that's going to bepublished Did it?
Was that the catalyst to thisnext paper, Exactly?
Canio (18:03):
Absolutely the way we, you know, whenever we do
research, let's say, even needsto resonate with the money and
the funding that we have.
So we need to be strategic inwhat we do because we need to
worth every penny, every dollarwe have.
So we need to move in thatdirection and every evidence
that whenever you publish apaper, you, whatever people read
(18:25):
, there are data.
But behind that paper there isa team of people that recently
talk, so they make briefing andmeeting, they analyze data,
contextualize data within theactual medical knowledge and the
future and the different otherscience.
How can I do that?
So it's everything youexperience coming from the paper
(18:45):
.
So that's why wheneverscientists publish paper,
they're excited.
Maybe nobody's going to read thepaper when it comes to people
like, not the medical community,but for us it's amazing because
behind that paper there areideas, confirmed, dismissed.
People get gut hungry, you know.
And then there is all the allof this stuff.
(19:06):
So there are humans, life,working for something and
challenging each other to dosomething different and change
the actual like management.
So it helped us to jump to theother program that actually is
not just this paper, but it isour history, other papers, like
they merged together and theyjumped in this new paper, that
(19:30):
is, a theoretical paper withsome computational predictions
or some preliminary data.
That is called PULSAR.
That's basically meaningprobabilistic and user-centered
learning for surgical adaptiverecommendation.
Alanna (19:46):
That's a long word for a lot of us, oh my gosh, alan.
Canio (19:50):
I think I could speak about that for hours, because
it's really what really drivesme.
So please stop me if I keepgoing like without any direction
.
But it's really something thatI, you know, whenever I go to
sleep.
Alanna (20:12):
I think about that, believe me.
So the research is like.
The words in the title itselfare more manpower than I want to
put in half the time just tothink of what it is.
What is this paper, though?
What does it entail?
Because I'm sure that it's likea new pathway to what the
future holds forpatient-centered care.
Canio (20:27):
Okay, Maybe we can talk about that by having a
conversation.
I want to try to introduce thethings by using metaphor.
Alanna (20:36):
Okay.
Canio (20:38):
So let's say that, do you like cooking, love it.
So whenever you cook, do youfollow a specific recipe, the
specific steps, or you go byyour mind?
A little bit of both.
Okay, exactly, exactly.
I love this answer.
So sometimes you change therecipe, you change ingredients.
The question here is how andwhat is the things that you
(20:59):
follow to change your mind aboutsome step.
You know what I mean.
Like, how do you?
We're talking about yourdecision-making process in your
mind.
So whenever you decide tochange something in your like,
let's say, management of thedish, what is the things that
(21:20):
really drive you?
Alanna (21:21):
If I like the flavor of something or if I don't have the
right ingredient, but somethingcan substitute it, so like
knowing what I have available tome versus what I want, if that
makes sense.
Canio (21:33):
Okay, it makes sense.
But you know, maybe you canopen your kitchen and you find a
hundred ingredients.
How do you pick especiallyspecifically that one instead of
another one?
Alanna (21:43):
What might work better in my mind, like what I feel
like would make it better.
Canio (21:49):
That's based on what your feeling is based on what?
Alanna (21:52):
Yeah?
Canio (21:52):
And what's your opinion is based on?
Alanna (21:54):
The way I feel or the way I prefer things.
Exactly yeah.
Canio (21:59):
Now we are entering what we call in this paper the gray
area dilemma.
So there is a gray area in ourmind of decision-making process,
especially in surgery, whereactually we do something.
Most of the process isevidence-based, but sometimes
during this process there arelittle choices in every surgery.
(22:20):
Every surgery is completelydifferent than anyone else and
from anyone else, because everypatient is completely different
than each other.
So then, surgery is a specificfield of medicine because it's
the same time diagnosis andtreatment, because you don't
have all the data that youreally need before the surgery.
Most of the time you discoversomething when you open the
(22:42):
patient and when you're thereyou're making diagnosis and you
have to change, sometimes, yourmind.
That's why, when we shareinformed consent, there is a
huge list of something that canhappen, like we are doing.
A surgeon that maybe you knowfor endometriosis they say I can
resect the bowel, you can comeup with a stoma, we can resect
ureter, it may be possible thatwe have a urinary stoma, you
(23:03):
know, but there is even a riskof fatality.
That's because we don't have aclear idea what's happening 100%
before opening the patient.
When you are there, you reallyhave the.
You know the reality, so youneed to adapt your
decision-making process as asurgeon between evidence coming
(23:24):
from science and then somethingthat you have in your mind
that's based on your experience,your intuition, something we're
still not able to identify.
That's the gray area dilemma,and the best surgeon is the one
that somehow has that gray areathat is much more efficient and
(23:44):
effective when, at the end, ittakes the choice for the patient
during surgery.
It doesn't mean that it'snecessarily the one that is the
oldest one, but it's the onethat the sum of his option is
actually the result as a result,as the best outcome for the
patient is actually the resultas a result, as the best outcome
for the patient.
(24:05):
And that's why surgery isextremely hard to teach, rather
than compared to any other fieldof medicine, because it's hard
to explain, objectify andstructure the gray area of the
decision making of the surgeon.
Now, where this study, you know,fit, this study fit in, uh in
in all of this.
So the thing here is I needsomeone, since, starting from
(24:26):
the previous study where wewanted to try to codify the uh
decision-making pattern of themachine, here is trying to
codify the decision-makingprocess of the human by using AI
.
The goal is using AI technologyUh, I'm not going to go into
the details in this but thehuman by using AI.
The goal is using AI technology.
I'm not going to go intodetails in this, but the goal is
using AI in order to detect andto study how the surgeon
(24:50):
behaves in all the differentscenarios, all the different
surgeries.
But the most important thing isthen to anchor and to attach to
this reasoning and evaluationwhat's happening to the patient,
because we need to anchor thealgorithm to the outcome for the
(25:11):
patient, because the matrixhere is I can judge whether the
action of a surgeon was betterthan the other one by relating
his action or her action to theoutcome of the patient.
And it's not just that, becausethen any patient is different.
Someone could say exactly thegoal is try to destroy them.
(25:33):
Let me say that the patient inall little pieces.
Those are data before thesurgery, with all the imaging,
objective evaluation, clinicalexamination, and so we get some
of the data.
Then, during the surgery, wehave a video, so we get
information coming from thevideo getting during the surgery
(25:54):
, and that's another package ofdata.
The other thing that's veryinteresting is that whenever you
do surgery, you are modifyingthe anatomy.
You are modifying the biologyof the body, so what's happening
is that we need to be able torelate our action to the
modification in that specificpatient.
Science is always working onunderstanding the differences,
so all of this little actionwill generate differences
(26:17):
between the patient, and so atthe end you have to like try to
picture that at one patientduring all the process, so
assessed in a dynamic way, sopre, during surgery, after
surgery, will be actuallyrepresented by billions of data.
And these data are actuallybelonging they're not like
random they are belonging tospecific compartments and that's
(26:40):
a called layer, and that'swhere actually multi-army comes
in.
So we don't just need to getthe data, but we need to
organize in their owncompartment and then we need to
be able to relate them in astructured way.
Each of them needs to be likepart of a compartment and then
we need to in this billion ofdata.
(27:02):
We need that to understand whatwas the strategy that actually,
in that specific case, helpedthe surgeon to get to the best
result for the patient.
And then in the next case, themachine would be.
After doing that, with differentphysicians and different people
, the machine would be able toguide us and also the patient,
(27:22):
by little by little, say okay,if you do this, be careful,
because last time you did that,you did the same stuff.
Although it's anatomically andtechnically correct, let me say
that was not good extremely goodfor the patient.
Because whatever we need to dois to anchor this to the patient
outcomes.
Most of the time if you talk tophysicians, to surgeons, they
(27:44):
say I did this and it wasperfect.
The other one did another stuffthat was indeed perfect.
So here is not about doing whatphysician or surgeon says is
perfect, but is to anchor any,every single step to the patient
outcomes.
And that is going to improveeven understanding of how a
surgeon thinks.
(28:05):
In that gray area is evenmedical education and also in
the shared decision making,because at this way, after some
times, before the patient comesto me, you know, before the
surgery I can, by looking at thedata, say okay, that's the best
choice for you among the allpossible ones and it's going to
give you that specific risk andthat specific benefit.
Alanna (28:28):
Interesting.
So it's like a, it's a guidefor the doctor to give the
patient the best outcome whilealso giving, like the patient, a
voice in the long-term outcome.
Exactly, exactly.
That's great and that's likesomething, because I think we
always say it's not aone-size-fits-all, especially
with endometriosis surgery, andso just tailoring the approach
(28:52):
to the actual patient prior tosurgery for better outcomes
long-term, is that kind of Wow,I was so confused when I
explained all of this and yougot everything very clearly.
Canio (29:05):
I don't know how you did it.
I was so.
It was a mess when I said that,because I follow my inner flow,
you know what I mean.
Alanna (29:12):
No, I picked it up.
I just am like fascinated bythe fact that you know, as
patients, we are constantlylooking at okay, like, is this
bowel resection right for me, oris this whatever resection
right for me, or is thiswhatever Is it right for me?
And this is just a good, moreeducated way of approaching a
surgery and making it trulymultidisciplinary from, like,
(29:33):
the AI standpoint, mixed withthe provider, mixed with the
patient, the outcome beingbetter.
This is crazy to me.
Canio (29:41):
What you said.
See, this conversation isbeautiful for this reason,
because it's I'm learningsomething from you and I or
treatment, we take our decisionby looking at data coming from.
We say, population study, soall the data we have.
(30:15):
We say, ok, you have to do thatbecause the chance of success
is this or because it's best foryou.
But OK, the question is how, asa patient, how am I similar to
the population that was using inthe study that generate the
evidence?
And that's where medicine isnot precise, because the
evidence are coming from a setof data that not necessarily are
(30:38):
the same of the very nextpatient that is in front of me.
And when it comes to surgery,it's even more difficult,
because I don't give the samepills to everybody, but actually
I'm tailoring the surgicalprocedure to the specific
patient and that's unique in anypatient.
So, whatever you introduce adecision and the differences,
you are making thedecision-making process much
(31:01):
more complicated.
Alanna (31:02):
Yeah.
Canio (31:03):
And very hard to understand.
Alanna (31:04):
Is this what?
When you did this paper, werethey seeing video or AI like
virtual reality ways of doingthe surgery, or was it just
words Like was it just feedback?
Canio (31:16):
through.
That's a very, very interestingquestion because it's like it's
actually a very detailedquestion and so I need to jump
necessarily to details.
So in order to design, first ofall, you start by designing the
algorithm that is moreefficient, Because in order to
(31:37):
do something, you can do it indifferent ways.
Then you have to find the onethat really is tuned in the
right way.
So this starts as a theoreticalstudy coming from observation
that I had personally in my lifeexperience as a physician and
scientist, Because one of thefantastic things you know, if
(32:00):
you really want to judge I don'twant to say judge, but if you
really want to have the realfeeling of a surgeon, I think
it's based on how really hetraveled in his life.
So because as much she has orshe has been exposed to
something different, then he canunderstand really what's the
best things.
(32:20):
Because whenever you are in thebest center in the world,
they're going to teach youthat's the right method and
that's okay because it comesfrom a school.
But you need to get to thatconclusion after you see
different methods and you startto believe actually, that is to
see that there are surgeons thatare doing different kind of
techniques and they still havetreating in the best way
(32:42):
possible and patients are happyabout that.
So the thing is coming from twoproblems.
So evidence, real problem,where you see all of these, a
surgeon in school says that theyare doing the best things for
the patient.
And then methodological problem.
That's coming from my researchfield, where actually you have
to question everything you do.
And how do you stop questioningwhat you do when you have no
(33:05):
more grade zones?
How do you not have grade zones?
You start to get the data andstart to, mathematically
speaking, give a system for eachof the data you come in game.
So the question is what kind ofdata are there?
The kind of data, every kind ofdata.
So they comes from imaging,they come from talking, so word
(33:26):
they come from.
It's called finite elementanalysis.
So what this conceptualizationthinks about is that we get the
data from ultrasound or RMI andwe translate those data that are
basically pixel in a system, acomputer that is able to
(33:46):
simulate the real functioning ofthe anatomy, but in a
functional way, so you cansimulate what's happening.
You say, okay, I want to dothis procedure.
You simulate the procedure andyou see that something else is
happening and that's why you cantailor the surgery before doing
the surgery, and then it'sgoing to tell you that the
(34:07):
things you're doing is good.
Well, this is easy when it comesto cancer.
So oncological surgery, becausethe goal is overall survival.
But when it comes to function,it's a completely different
stuff, Because how do you reallyassess functionality in women
today?
By some questionnaire what isfunctionality?
It's actually really.
It's different for every woman.
(34:28):
So, for example, I can say thatfor a young athlete woman,
functionality is being able tohave no pain and to be as much
as she can and give the bestperformance.
For an old lady, maybe thefunctionality is able to hold
(34:50):
the hands of her grandson.
So even if she's pained, it'snot a problem.
So is everything based on thisquestion?
How can I representmathematically people's problem
and human's problem?
Because whatever you aresearching, you're human acting
and not human.
And so when the study will bepublished, the cool things are
(35:14):
going to be the architecture ofall of this, because it's going
to be extremely detailed andevery step is connected to
another one.
Alanna (35:22):
And then there are loops that reinforce and retune all
the system again connect it toanother one, and then there are
loops that reinforce and retuneall the system again.
It's like taking both thevisual of the scans and then
you're taking the patient.
Does it account for, likepatient symptoms?
Does it account for, like allof the functional things that
are at that point failing themfor lack of a better word and
(35:42):
taking that into considerationand giving you or the provider
the approach that's going tobest suit them long term in a
visual manner?
Or is it just like spelling itout, it's like a checklist, or
is it like a visual aid for them?
Canio (35:59):
And that's where it comes .
Another project, right, andthat makes me feel that we are
going in the right direction.
This is something very, very,you know, exclusive, preliminary
, because here comes the secondpoint communication.
And do you really understandwhether what you're telling
(36:22):
people patient you're reallyable to explain in a clear way?
Are you able, really you know,whenever you say the chance of
success is 80% and the failure20%?
Okay, and the patient said thatevery patient will believe
they're going to be in the 20%.
So there is a problem here, evenin communication, and that's
(36:43):
another part of the problem thatis not focused in this paper,
but it's another project we areleading, that's translational
communication in the medicine,and we are working together with
visual artists, so they arehelping us in trying to
understand how to translatecomplicated medical stuff in
(37:04):
something that people can reallyunderstand, but not logically
understand, even emotionallyunderstand.
They need to understand that,because when you say 80, 20% are
very, you know, great data, butthe 80% is not just the 80%,
it's also what's happening ifyou don't do that kind of
(37:25):
procedure.
So if you don't do that kind ofprocedure, so if you don't
choose that procedure, somethingelse is going to happen.
So it's giving the right let'ssay even emotional way to pure
great data?
The answer for me I don't know.
I'm working together with thevisual artists, so people that
work with that are inspired byemotion much more than me, that
(37:50):
have been exposed in life tosomething different than me.
So in science and that's thebeauty of the United States, I
have to tell this Whenever youwork in university, you have all
of these academic peopleworking together and give their
point of view and really, at theend, try to do something that
is really effective, not justbeautiful chrome.
It needs to work out.
Alanna (38:10):
Right, I'm excited to see where that one goes, because
I feel like that could changethe narrative of outcomes.
Canio (38:17):
Yes, which?
Alanna (38:17):
is exciting.
That's a wrap for EndoBattery'sFast Charge this week.
I hope this episode left youinspired and empowered to
continue advocating for yourcare and encourage that research
is happening and change canhappen in our lifetime.
Make sure to join us next timeas Kanyo sits down with us to
explore more research that'sbeing done.
(38:39):
Until then, continue advocatingfor you and for others.
Popular Podcasts
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.