AI in Healthcare: The Next Frontier | Leonardo Castorina | TEDxUniversityofEdinburgh
Summary
TLDRCe script explore l'application de l'intelligence artificielle (IA) dans le domaine de la santé, soulignant les défis tels que la résistance aux antibiotiques et les nouvelles maladies. Il explique comment l'IA, en analysant des données variées, peut aider les médecins et les infirmières à améliorer la prise en charge des patients, réduire les temps d'attente et traiter des problèmes tels que la fatigue des professionnels de la santé. Des exemples concrets illustrent les avantages de l'utilisation de l'IA pour la préparation aux consultations, la prise de notes et la recherche médicale, ainsi que pour l'analyse d'images médicales. L'auteur met en garde contre les erreurs potentielles de l'IA et souligne l'importance de la sécurité, de la transparence et de la confidentialité des données.
Takeaways
- 🚑 Le secteur de la santé fait face à de nombreux défis, tels que la résistance aux antibiotiques et l'apparition de nouveaux virus.
- 🤖 L'IA est présentée comme un outil magique qui peut changer la technologie et pourrait avoir des applications dans le domaine de la santé.
- 💾 L'IA commence avec des données, qui peuvent être du texte, des images ou d'autres types, et utilise un système d'apprentissage pour identifier des motifs.
- 👨⚕️ L'IA peut aider les médecins et les infirmières en réduisant le temps de consultation et en améliorant la gestion des dossiers médicaux.
- 📚 L'IA peut aider à réduire la charge administrative des médecins en automatisant la prise de notes et la maintenance des dossiers médicaux.
- 🔍 L'IA peut être utilisée pour la recherche, en identifiant de nouvelles recherches scientifiques ou en analysant les symptômes des patients.
- 📈 L'IA peut aider à améliorer la résolution des images médicales, telles que les IRM, en augmentant la qualité des scans à partir de scans de faible résolution.
- 👁️ L'IA a été utilisée pour analyser les scans de la rétine, permettant d'identifier si un patient a besoin d'une consultation urgente ou non.
- 🧬 L'IA peut être utilisée pour l'étude des protéines, qui sont essentielles à la vie et peuvent être utilisées pour développer de nouveaux traitements.
- 🛡️ Il est important de considérer les problèmes potentiels liés à l'IA, tels que le biais, la transparence, la confidentialité, la responsabilité et la sécurité.
Q & A
Quelle est la crise de santé mentionnée dans le script?
-La crise de santé mentionnée est caractérisée par le surmenage des médecins et des infirmières, l'allongement des délais d'attente et de nouveaux défis tels que la résistance aux antibiotiques et l'apparition de nouveaux virus.
Quel rôle l'IA pourrait-elle jouer dans le domaine de la santé?
-L'IA pourrait aider les médecins et les infirmières en réduisant la paperasserie, en améliorant la couverture médicale en raison du manque de personnel et en aidant à prédire des tendances ou des diagnostics en analysant des images médicales ou des données de santé.
Quels sont les types de données que l'IA analyse dans le contexte de la santé?
-L'IA analyse divers types de données, tels que les dossiers médicaux textuels, les images telles que les radiographies ou les scans, et même des données de marchés comme les fluctuations du marché boursier.
Que signifie le terme 'généralisation' dans le contexte de l'IA?
-La généralisation fait référence à la capacité d'un système IA formé de reconnaître les mêmes motifs dans de nouvelles données non vues auparavant, comme identifier un cancer dans une nouvelle radiographie de thorax.
Pourquoi les médecins éprouvent-ils du burn-out?
-Les médecins éprouvent du burn-out en raison de la paperasserie et du manque de personnel, ce qui signifie qu'ils doivent gérer un nombre important de tâches administratives et médicales en plus de soigner les patients.
Quelle est la différence entre les médecins et les infirmières en termes de vacaturs?
-Pour les médecins, sur 20 postes vacants, seules 19 sont pourvus, tandis que pour les infirmières, sur 10 postes vacants, seulement 9 sont pourvus, ce qui répartit le travail non effectué sur les infirmières et les médecins actifs.
Comment l'IA pourrait-elle aider à réduire les temps d'attente dans les hôpitaux?
-L'IA pourrait aider à réduire les temps d'attente en automatisant certaines tâches telles que la préparation des dossiers médicaux, en aidant à prendre des notes lors des consultations ou en identifiant rapidement des cas d'urgence à partir d'images médicales.
Quels sont les trois principaux domaines d'application de l'IA en médecine mentionnés dans le script?
-Les trois principaux domaines d'application de l'IA en médecine sont la préparation avant une rencontre avec un patient, la prise de notes lors des consultations et la recherche médicale.
Quels sont les problèmes potentiels que l'IA pourrait rencontrer dans le domaine de la santé?
-Les problèmes potentiels mentionnés sont le biais, la transparence, la confidentialité, la responsabilité, la rétroaction de l'utilisateur et la sécurité.
Pourquoi est-il important de ne pas remplacer les médecins par l'IA, mais plutôt de les aider?
-Il est important de ne pas remplacer les médecins par l'IA car les modèles IA ne sont pas parfaits et ont un taux d'erreur qui n'est jamais de 0%. Les médecins sont nécessaires pour prendre des décisions éclairées et pour intervenir si l'IA fait une erreur.
Quel est l'exemple donné pour montrer comment l'IA pourrait aider dans la recherche de nouvelles thérapies?
-L'exemple donné est l'utilisation de l'IA pour concevoir une nouvelle forme d'anticorps qui pourrait s'attacher à la protéine pic de la COVID-19, réduisant ainsi le temps nécessaire pour trouver un traitement ou un vaccin.
Outlines
🏥 L'Intégration de l'IA dans le domaine de la santé
Le premier paragraphe aborde la crise actuelle du secteur de la santé, où les médecins et les infirmières sont surmenés et où les temps d'attente augmentent. Il évoque les défis tels que la résistance aux antibiotiques et l'apparition de nouveaux virus. L'orateur explique l'intérêt de l'utilisation de l'intelligence artificielle (IA) dans ce domaine, en soulignant qu'elle peut aider à analyser des données médicales, tels que des dossiers médicaux et des images médicales. L'IA est décrite comme un outil de reconnaissance de motifs qui s'améliore au fil du temps grâce à un processus d'apprentissage. L'orateur mentionne également l'importance de la généralisation, qui permet à l'IA de reconnaître des motifs dans de nouveaux cas. Enfin, il cite des exemples d'interactions avec l'IA, comme les assistants vocaux et les voitures autonomes, et soulève la question de la confiance en l'IA.
🤖 L'IA au quotidien et son application dans les soins médicaux
Le deuxième paragraphe traite de l'épuisement professionnel des médecins, souvent dû à la paperasserie et à un manque de personnel. L'orateur présente l'IA comme un outil pouvant aider à réduire cette charge de travail. Il explique comment l'IA peut être utilisée pour préparer les consultations, en analysant les dossiers médicaux des patients, pour documenter les consultations en temps réel et pour faire la recherche médicale. L'orateur mentionne également l'utilisation de l'IA pour traiter des images médicales, comme les IRM, en améliorant la résolution des images et en aidant à identifier les anomalies. Il cite également l'exemple de Google DeepMind, qui a utilisé l'IA pour analyser des scans rétiniens et déterminer l'urgence des soins. L'orateur souligne que l'objectif de l'IA n'est pas de remplacer les médecins, mais de les aider, tout en reconnaissant les erreurs potentielles et en proposant des solutions pour y faire face, telles que la réduction des biais, la transparence, la confidentialité des données, la responsabilité et la sécurité.
🧬 La prochaine avancée de l'IA : les protéines
Le troisième paragraphe se concentre sur l'utilisation potentielle de l'IA pour la découverte de nouvelles protéines, qui sont essentielles à la vie et à la santé. L'orateur décrit les protéines comme des 'architectes de la vie', puisqu'elles sont impliquées dans presque tous les processus biologiques. Il explique comment l'IA pourrait aider à concevoir de nouvelles protéines ou à identifier des protéines existantes qui pourraient avoir des propriétés thérapeutiques. L'orateur donne l'exemple de la lutte contre le COVID-19, où l'IA pourrait être utilisée pour concevoir des anticorps spécifiques qui se lient à la protéine Spike du virus. Il souligne que l'IA peut accélérer ce processus et potentiellement révolutionner la recherche médicale. Cependant, il rappelle également l'importance de gérer les défis associés à l'utilisation de l'IA, tels que la sécurité et la fiabilité des algorithmes.
Mindmap
Keywords
💡AI
💡Cancer
💡Burnout
💡Understaffing
💡Waiting Times
💡Antibiotic Resistance
💡Generalization
💡Data
💡Training
💡Bias
💡Proteins
Highlights
Healthcare crisis with doctors and nurses burning out and increasing waiting times.
AI as a tool that can change technology and its potential applications in healthcare.
AI systems analyze data like medical records and X-ray scans to identify patterns.
Training AI systems to improve pattern recognition over time.
The importance of generalization in AI for handling new data.
AI's role in search engines to identify relevant websites.
AI's potential to assist doctors and nurses by reducing paperwork and understaffing issues.
Burnout among doctors due to paperwork and understaffing.
Job vacancies in healthcare, particularly the shortage of doctors and nurses.
Increasing waiting times for patients to see a doctor, exacerbated by COVID-19.
AI chatbots like ChatGPT can assist in healthcare by preparing for patient meetings, logging health records, and researching.
AI's ability to increase the resolution of low-quality MRI scans.
Google DeepMind's work on analyzing Retina scans to determine the urgency of patient referrals.
The importance of not replacing doctors with AI but instead assisting them.
Challenges AI faces in healthcare, such as bias, transparency, privacy, accountability, and safety.
The next frontier for AI: working with proteins and their potential in creating new cures.
AI's potential to design antibodies that bind to viruses like COVID-19, speeding up the process of finding cures.
The need for caution and responsible deployment of AI in healthcare.
Transcripts
we're in the middle of a healthare
crisis doctors and nurses are burning
out waiting times keep getting longer
and longer and we Face new challenges
such as antibiotic resistance and new
viruses you might have heard of AI as
this magical tool that's changing
technology and you might have thought
about Ai and Healthcare and whether
there's some applications uh so I'm
going to show you a few applications and
why I'm really excited about this now
you might have already interacted with
AI in some way maybe with Siri Google
Alexa smart devices or maybe you've seen
or been in an autonomous car cool uh or
maybe you just watchh Terminator and you
think we should burn all this AI
business to the ground so maybe we
should start asking ourselves what is AI
and why should we use it in
healthcare no what is
AI so we first start with data and data
can come in the form of text such as
medical record x-ray scans such as
images and just data in the form of
whatever uh sound stock market all of
this data gets analyzed by this AI
system which is just a pattern
recognition tool we're just looking at
the data and seeing what's in
there the system under goes a process
called training essentially what happens
is that the model gets better and better
and better at identifying these patterns
now you will see that it never actually
reaches 100% And if it does in my
experience there's usually something
wrong with
it so once you have a trained AI system
you hope to have something we call
generalization generalization is when
you have new data coming in and your AI
system is able to identify the same
patterns for example if a new patient
comes in your hospital and you get a
chest x-ray then you're able to
understand whether the patient has
cancer or not or maybe Elon Musk goes on
a Twitter run page and we can predict
that the stock market uh price of Tesla
is going to go down and finally search
engines you probably have interacted
with search engines that identify the
the websites uh that you're most
interested
in this is all cool we can see how AI
can be used in Tech now how about AI in
healthcare and why should we use it the
short answer here is to help doctors and
nurses and I have an example for this
pajama time now for most of us pajama
time literally just means to get ready
for sleep unfortunately this is not the
case for doctors for doctors pajama time
means completing the tasks that they
couldn't do during the day and this
includes um medical records maintaining
them um doing bureaucratic work for the
hospital or simply keep up with the
research and it's no wonder that
actually two-third of doctors mentioned
burnout as the main cause of distress
now there are two main causes of burnout
for doctors one is paperwork two under
Staffing let's start with paperwork as I
said there's a lot of paperwork involved
when you have patients you need to make
sure that you take notes when the
patient is telling you something
important and you need to make sure that
you are
listening and the second one is under
Staffing and for this I have an example
about Job vacancies now the data is only
for England unfortunately I couldn't not
find the data for
Scotland
now we have job vacancies for doctors so
for every 20
doctors only 19 of them are filled so
that means that there's one doctor job
vacancy that that the that is
redistributed across the 19 of the
doctors if you thought that was bad
nurses for every job uh every 10 job
vacancies for nurses only nine of them
are filled that means that the work of
that one nurse is spread across the nine
nurses and this is just the tip of the
iceberg now let's look at waiting times
again NHS England sorry Scottish people
in the audience um you can see the this
is the number of patients on the y axis
for the last um I don't know how many
years essentially it goes up and it goes
up to a point um where we're getting
close to 8 million people and this has
gotten worse especially after Co now for
your contact
there are 50 million people in England 5
Z so that means there's about one in
five people that are waiting to see a
doctor this is insane and we have an
aging population so this is only going
to get
worse so how can AI help essentially you
might have heard of this um chat B
really cool guys that was not even a
joke you might have heard of chat GPT
now is a good friend of mine uh if you
don't know what it is is a smart chat
chatbot you can ask it questions it will
reply um hopefully with the right answer
um but you can ask it to do various
things like coding you can help it write
um you can help it ident can help you
identify new scientific
papers now I thought it'd be cool for
Chad GPT to say hi to all of you before
we start talking about it and I believe
this is the first time in tedex history
that this happened so
yeah so I think there are three main
ways Chad GPT can help or technology
thereof one is for Preparation so before
you go into a meeting with a patient CH
GPT analyzes all the health records and
tells you well you should follow up on
these specific
things second one is logging and here I
mean logging in the health record so
while you are with a patient uh we want
the doctors to spend as much time with a
patient as possible we don't want them
to be taking notes uh we want them to to
feel heard so one way we can do this is
to automatically record a conversation
and then have a system that summarizes
the main points of the conversation and
the third one is research as I said is
extremely hard keeping up with the
research it is for me and even harder
for doctors the cool thing about Chad
GPT is that it interacts with us with
language so if there's a new symptom
that comes up in the patient record it
can identify new scientific research
that shows maybe new diseases correlated
with that
symptom so we talked about language now
let's talk about
images this is an MRI scan and actually
this is work that I was involved in so
when you get an MRI scan you get into
this machine and you're you should be
very very still you're um because if you
move even slightly your scan will be
blurred like the one over there now you
can't do much with a blurred scan so the
doctor can either try their best to
identify patterns whether you have brain
cancer or anything or you will need to
do it again now this is extremely
uncomfortable as I said so what we did
is we took low resolution scans brain M
scans and we took high resolution ones
and we taught an AI system to
automatically increase the resolution of
these scans and now from this scan what
you can get is something like this now
this this you can get it with the
computational power of your phone so
it's very fast it's very easy to run and
saves you time uh saves doctor's time
saves patient time very cool my final
example
is Retina scans essentially when you go
to an optometrist they look at the back
of your eye and see if there's anything
wrong now if there is anything wrong
they will refer you for a surgery and
they will also attach an urgency to that
surgery what Google Deep Mind did is
based on the scans identify whether the
patient needed to be seen urgently or
Not by a healthcare professional and now
I have the data over here this is the
error rate on the referral decision you
have optometrists Retina Specialists and
AI now you can see that thei matches
more or less the the performance of the
of the the healthcare specialist so this
is very cool because again these models
run really quickly and they can be
deployed in areas where there just
aren't that many doctors you can use
them to pre-screen the population so
that these high at risk people can be
seen by by the doctors now it's
important here I'm not saying we should
replace doctors we should help
them you might have noticed that uh the
error rate was never actually 0% and I
mentioned this is a problem in AI so we
should think about what happens when it
makes mistakes now I've prepared a five
point non-exhaustive list of what
happens like the problems that we might
face and ways we can mitigate them so
the first one is bias we don't want the
bias in society to be
um spread around um so one way we can do
this is make sure that the populations
in the data are well represented and we
can do audits so different companies can
check that the models are that the
models that are deployed are working
working
well the second one is transparency and
now this is a really hard one because
it's really hard to open up the AI
system and understand why they made a
specific
decision and I think here the only real
solution is to look at the open source
code now open source code means that you
and me can access the code for free and
understand what happens under the
hood privacy of course we want our data
to be private so it needs to be
encrypted and we want data minimize a so
we don't need to collect all the data we
possibly can about a patient we just
need the data that we
need accountability so when it makes
mistakes there should be a person or an
institution that's responsible for those
mistakes and it should be user feedback
so if the AI makes mistakes we should
learn from
them and finally safety of course we
want something safe to be deployed
especially in healthcare so one problem
here is called hallucination and
especially if you used TBT before you
can see that it confidently tells you
something even though it's false and you
can argue with it and it's like no no no
it's false and continues so it it is a
big problem so we need to make sure that
when these models are deployed they're
fully analyzed and there should be
fallback strategies so if the AI system
fails there should be a human taking
over the
AI now this is all cool I talked about
Ai and all the possible applications now
i' like to talk about the next Frontier
and the next Frontier is
proteins that was no joke um it's
actually my PhD
project so you might think of proteins
as the the things that you eat before
you go to the gym but actually proteins
do pretty much everything in the natural
world you can think of them as Legos now
as a kid I used to love Legos because
oh yeah because as long as you have
these instructions then you can build
your own spaceship or whatever it's very
cool and much like that proteins are
made of these building blocks called
amino acids and you can put them in
Chains and when you do you create these
proteins I like to call proteins The
Architects of life on Earth because as I
said they do pretty much everything I'm
going to show you a few examples of
this you'll know plants plants do
photosynthesis so this big guy over here
is rubisco helps with photosynthesis
it's one of my favorite
proteins um this is DNA DNA needs to be
replicated and to replicate DNA we use
DNA polymerase over
there so what I'm trying to say here is
that as long as you have instructions
which come in the form of
DNA you have your building blocks which
are amino acids then you can build your
your spaceship or a protein now as my
final example I have this thing that you
should be familiar with covid-19
this thing sticking out you guessed it
it's a protein it's called a spike
protein and you might be also familiar
with this other protein called an
antibody now what happens with
antibodies is that they attach to the
spike protein and they initiates an
immune response now to do this we
actually need to create lots of
different random shaped antibodies until
we find the correct one matches the
shape of the co and we initiate the
immune response this takes time
a long time which is why it takes us
time to heal up from
diseases now believe it or not this is
actually a similar approach that the
pharmaceutical industry uses in finding
new cures essentially they have lots of
compounds and they dump them all at the
problem and try to see which compound
actually decreases the um the side
effects or cures the disease and this is
really time consuming so what I'm
proposing here is to use Ai and I have a
an example with Co so if you have a
spike protein like the one I showed you
earlier you can ask an AI to dream up
the shape on an antibody or like the
most probable shape on an antibody that
will bind
Co now you can ask it
to design based on that shape the actual
antibody so there we have it instantly
we generated a new antibody
that we know will bind Co and the cool
thing about this is that if we have new
viruses coming up then we can instantly
generate new antibodies for those
viruses as well cool now much like any
new technology we need to be careful we
need to make sure that we know how to
use AI before we deploy
it but I believe that once we'll be able
to do that it will change the world
thank
you
Browse More Related Video
How AI can improve childhood mental health care | Bryn Loftness | TEDxUniversityofMississippi
Why CoPilot Is Making Programmers Worse
Et si l'IA était un nouvel instrument pour les musiciens ? | Benoit Carré | TEDxTours
L'IA DÉTRUIRA-T-ELLE L'HUMANITÉ !?
L'Intelligence Artificielle peut-elle remplacer l'homme?
How DePIN can revolutionize the way AI works (With Jonathan Schemoul from Aleph.im)
5.0 / 5 (0 votes)