The future of AI in medicine | Conor Judge | TEDxGalway

Summary

The speaker reflects on his journey from a young boy with a stammer in a drama competition to a medical consultant and lecturer, emphasizing the importance of data in solving medical mysteries. He introduces multimodal AI, which processes diverse data types, as a potential solution to healthcare inefficiencies exacerbated by technology. Highlighting examples of AI in healthcare, he discusses the need for trust, explainability, and clinical trials in implementing AI safely, advocating for a compassionate approach that integrates AI with human expertise.

Takeaways

• 🎭 The speaker's connection to the stage began 26 years ago as a 12-year-old in a drama competition, playing a detective in a play about a hotel robbery.
• 🕵️‍♂️ The role of a detective in the play mirrors the speaker's current role as a medical consultant, both involving the collection of information to solve a mystery.
• ⏱️ The speaker spends 70% of their time collecting information and 30% making decisions, a ratio that has been exacerbated by technology in the healthcare field.
• 💼 The introduction of electronic health records, designed for billing rather than efficiency, has increased administrative workload for doctors, reducing patient interaction time.
• 🤖 The speaker introduces the concept of multimodal AI, which processes various types of data, as a potential solution to the information overload in healthcare.
• 👨‍⚕️ The speaker differentiates between single-model AI, like machine learning and computer vision, and multimodal AI, which can process text, images, and numbers.
• 🏥 Examples of single-model AI in healthcare include 'Chest Link' for X-ray triage, an AI model for diagnosing eye diseases and predicting outcomes, and another for predicting Parkinson's disease from retinal images.
• 🗣️ Google's 'Med-PaM' is a large language model that has passed the US medical licensing exam, demonstrating the potential of AI in medical question answering.
• 📲 The multimodal version of 'Chat GPT' and 'Med-PaM M' can analyze medical images and text, providing insights that sometimes surpass human radiologists' reports.
• 🛡️ For multimodal AI to be implemented safely, trust, explainability, and randomized clinical trials are essential to ensure its effectiveness and reliability.
• 🤝 The future of healthcare with AI should prioritize compassion and understanding, allowing doctors to spend more time with patients and improving health outcomes.

Q & A

What was the speaker's first experience on stage 26 years ago?

The speaker's first experience on stage was at the town hall theater, participating in a drama competition for local schools where he played a detective in a play written by his best friend.

How does the speaker compare his role as a detective to his current profession?

The speaker compares his role as a detective solving the mystery of who robbed a hotel to his current profession as a medical consultant and lecturer, where he solves medical mysteries by collecting and analyzing patient data.

What is the 70/30 imbalance the speaker refers to in healthcare delivery?

The 70/30 imbalance refers to the speaker spending 70% of his time collecting information about patients and only 30% making decisions and communicating with them. This imbalance is common in healthcare and exacerbated by technology.

What is the main idea the speaker wants to share about artificial intelligence (AI) in healthcare?

The speaker wants to introduce the responsible use of medical AI, specifically multimodal AI, as a potential solution to improve healthcare efficiency and reduce administrative workloads, allowing more time for patient care.

What is multimodal AI and how does it differ from single model AI?

Multimodal AI can process and integrate different types of data such as text, images, and numbers, similar to how a doctor gathers various types of information from a patient. Single model AI, on the other hand, usually processes only one type of data at a time.

What are some examples of single model AI in healthcare mentioned by the speaker?

Examples include OxyPit’s chest link for triaging chest x-rays and an AI model from University College London that diagnoses eye diseases and predicts conditions like Parkinson's disease from retinal images.

What is the significance of the Med-PaLM medical AI model mentioned by the speaker?

Med-PaLM is a large language model trained to perform medical question answering, and it is notable for being the first AI to pass the US medical licensing exam with a passing score, demonstrating its potential in assisting medical professionals.

How does the speaker envision the future use of multimodal AI in healthcare?

The speaker envisions multimodal AI being used to enhance healthcare delivery, especially in remote and underserved areas, by integrating various types of patient data to provide insights and support for medical professionals.

What are the key factors necessary for implementing multimodal AI safely in healthcare?

The key factors are trust, explainability, and randomized clinical trials. Trust involves patients feeling comfortable with AI in their care, explainability ensures understanding of AI decisions, and clinical trials test AI models like medicines to ensure effectiveness and safety.

What is the 'eyeball test' and why is it important according to the speaker?

The 'eyeball test' refers to the initial visual assessment of a patient by a medical professional. It's important because it provides context to diagnostic results and has been shown to be highly accurate in emergency settings. The speaker emphasizes that even with AI, the human element remains crucial.

Outlines

00:00

🎭 From Young Detective to Medical AI Advocate

The speaker reflects on his journey from performing in a school drama competition 26 years ago to his current role as a medical consultant and lecturer. He draws a parallel between his past role as a detective and his present role in diagnosing patients, emphasizing the significant time spent on data collection in both scenarios. The speaker also highlights the challenges posed by technology in healthcare, such as the electronic health record system, which has increased the administrative workload for doctors, thereby reducing the time available for patient interaction. He introduces the concept of multimodal AI as a potential solution to these challenges, explaining that unlike single-model AI, which processes one type of data, multimodal AI can handle various forms of data, similar to human intelligence in a clinical setting.

05:01

🤖 Cutting-Edge Examples of Single-Model AI in Healthcare

The speaker provides three examples of single-model AI applications in healthcare, demonstrating their capabilities and potential impact. The first is 'Chest Link' by OxyPit, an AI system that autonomously triages chest X-rays for abnormalities, only involving human radiologists when abnormalities are detected. The second example is an AI model developed at University College London that can diagnose eye diseases and predict outcomes from conditions like macular degeneration, as well as predict Parkinson's disease from retinal images. The third example is Med-PaM, a medical large language model by Google that has passed a U.S. medical licensing exam, showcasing the potential for AI in medical question answering. The speaker emphasizes the importance of using AI in conjunction with healthcare professionals and introduces the concept of multimodal AI, such as the multimodal version of Chat GPT and Med-PaM M, which can process various types of medical data and perform multiple medical tasks.

10:04

🛡 Implementing Multimodal AI Safely in Healthcare

The speaker discusses the necessary components for the safe implementation of multimodal AI in healthcare, including trust, explainability, and randomized clinical trials. He cites a survey indicating public anxiety about healthcare workers relying on AI and the fear of rapid integration without fully understanding the risks. The speaker stresses the importance of explainable AI to demystify the decision-making process of AI models, allowing healthcare professionals to understand and potentially challenge the AI's recommendations. He also advocates for randomized clinical trials to test AI models, comparing their effectiveness to traditional methods. The speaker concludes by emphasizing the importance of compassion and understanding in the relationship between AI and healthcare professionals, envisioning a future where multimodal AI can make healthcare more efficient, personalized, and accessible, especially in remote areas with limited access to specialized care.

Keywords

💡multimodal AI

Multimodal AI refers to artificial intelligence systems that can process and analyze data from multiple sources, such as text, images, and numbers. In the context of the video, the speaker discusses how multimodal AI can enhance healthcare by integrating various types of medical data to improve diagnosis and patient care. An example given is the use of different types of input like chest x-rays, skin images, and pathology reports.

💡electronic health record

An electronic health record (EHR) is a digital version of a patient's paper chart, containing comprehensive health information. The video highlights how EHRs, designed mainly for billing purposes, have increased the administrative workload for healthcare professionals, reducing the time they spend with patients. This imbalance is one of the issues that multimodal AI aims to address by streamlining data collection and analysis.

💡triage

Triage is the process of determining the priority of patients' treatments based on the severity of their condition. In the video, the speaker describes how AI, like the chest x-ray triage system by Oxipit, can efficiently identify normal and abnormal x-rays, allowing radiologists to focus on cases that require immediate attention. This task-sharing between AI and human professionals exemplifies how AI can enhance medical workflows.

💡large language model

A large language model is a type of AI designed to understand and generate human language by processing vast amounts of text data. The video mentions ChatGPT, a large language model by OpenAI, and Google's Med-PaLM, which has been trained to pass medical exams and assist in medical question answering. These models illustrate the potential of AI to assist healthcare professionals in diagnostic and educational tasks.

💡explainable AI

Explainable AI refers to artificial intelligence systems that provide understandable and transparent explanations for their decisions and outputs. The video emphasizes the importance of explainable AI in healthcare, where understanding why an AI model recommends a particular treatment is crucial for trust and safety. This transparency helps bridge the gap between AI recommendations and human decision-making.

💡randomized clinical trial

A randomized clinical trial (RCT) is a scientific study that tests the efficacy and safety of medical interventions by randomly assigning participants to different treatment groups. The speaker stresses the need for AI models to undergo rigorous testing through RCTs to ensure their effectiveness and reliability in real-world medical settings. This methodical approach is essential for integrating AI safely into healthcare practices.

💡Parkinson's disease

Parkinson's disease is a progressive neurological disorder that affects movement, causing symptoms like tremors and difficulty walking. The video discusses an AI model developed at University College London that can predict Parkinson's disease years before symptoms appear by analyzing images of the retina. This capability demonstrates AI's potential to detect diseases early, even when traditional diagnostic methods may not.

💡confirmation bias

Confirmation bias is the tendency to search for, interpret, and remember information in a way that confirms one's preconceptions. In the context of AI in healthcare, the video warns about the risk of doctors uncritically accepting AI recommendations that align with their existing beliefs. Explainable AI can help mitigate this risk by providing reasons for its conclusions, encouraging more informed and balanced decision-making.

💡task-sharing

Task-sharing involves distributing specific tasks between humans and AI to optimize efficiency and accuracy. The video provides the example of AI triage systems that handle routine x-ray evaluations, freeing up radiologists to focus on complex cases. This collaborative approach enhances healthcare delivery by leveraging the strengths of both AI and human expertise.

💡compassionate care

Compassionate care in healthcare involves understanding and addressing patients' emotional and psychological needs, not just their physical symptoms. The video contrasts the capabilities of AI with the human touch, emphasizing that while AI can assist with diagnostics and data analysis, it cannot replace the empathy and personalized care provided by human healthcare professionals. This balance is crucial for holistic patient care.

Highlights

The speaker's first experience on stage 26 years ago as a 12-year-old boy with a speech impediment in a drama competition.

The parallel between the speaker's past role as a detective in a play and his current role as a doctor solving medical mysteries.

The common 70/30 ratio of time spent by healthcare professionals on data collection versus decision-making and communication.

The impact of electronic health records on reducing the time doctors can spend with patients due to increased administrative workload.

The introduction of multimodal AI as a potential solution to improve healthcare efficiency.

Definition and explanation of multimodal AI that processes various data forms like text, images, and numbers.

The release of Chat GPT by Open AI and its significance in the AI field.

Examples of single-model AI in healthcare, such as the chest X-ray analysis software Chest Link by OxyPit.

The AI model developed by University College London for diagnosing eye diseases and predicting outcomes.

The capability of the same AI model to predict Parkinson's disease from retina images, years before symptoms appear.

The importance of using AI in conjunction with healthcare professionals for diagnosis and care.

Google's release of Med-PaM, a medical large language model that passed the US medical licensing exam.

The demonstration of multimodal AI's ability to analyze an ECG and provide medical advice.

The release of Med-PaM M by Google, capable of handling multiple medical tasks with various input types.

The need for trust, explainability, and randomized clinical trials in implementing multimodal AI safely.

Survey results showing public anxiety about healthcare workers relying on AI for medical treatment.

The concept of explainable AI to understand the reasoning behind AI model decisions.

The necessity of randomized clinical trials for AI models to ensure their effectiveness and safety.

The importance of the 'eyeball test' in medicine and the potential for integrating patient images or videos into multimodal AI models.

The vision of using multimodal medical AI to make healthcare more efficient, personalized, and accessible globally.

The emphasis on prioritizing compassion and understanding in the integration of AI with human healthcare professionals.

Transcripts

[Applause]

[Music]

the last time I stood on this stage in

the town hall theater was 26 years

ago I was a handsome 12year old

boy I was in a drama competition for

naal schools in a play written by my

best

friend in that play I was a detective

trying to solve a mystery of who robbed

a fictional Hotel the hotel was called

Hotel El chipo

Nono I was also a boy with a stammer or

a speech

impediment and I was desperately trying

to both remember and say my lines

during that play I spent 70% of my time

collecting information to solve this

mystery of who robbed the

hotel so fast forward 26 years and not

much has

changed I'm now working as a medical

consultant in the hospital for half of

my time and as a senior lecturer in

applied clinical data analytics in the

university for the other half the

context has changed from a detective

solving a mystery of who robbed a hotel

to a doctor solving a

mystery what is the cause of illness in

this patient in front of

me and I still spend 70% of my time

collecting information about the patient

and only 30% of my time making decisions

on that information and communicating

with the

patient so the information or data that

I collect takes many forms patients

blood pressure their medical history

blood test

results and this inbalance in how

Healthcare is delivered this

7030 is well known all over the

world and in many specialities it's been

made worse by technology with the

introduction of the electronic health

record for example that was designed for

collecting information about billing and

not designed to to make Healthcare more

efficient so this extra administrative

workload that doctors have to do reduces

the FaceTime that they have with

patients the FaceTime that we were

fundamentally trained to do and the

FaceTime that the patients

want so the idea worth spreading that

I'm going to share tonight is a

potential solution to

this we've all heard a lot about the

risks of AR artificial

intelligence but I want to introduce A

New

Perspective one where the responsible

use of medical AI could help to solve

some of these

problems and I'm going to introduce a

new type of AI called multimodal ai

multimodal ai is AI that takes data in

many different forms text images

numbers when I work as a doctor in the

hospital I'm talking to the patient I'm

listening to the patient I'm listening

to their chest with a stethoscope I'm

palpating their abdomen I'm looking at

their blood test results this is

multimodal human

intelligence multimodal meaning lots of

different types of

data so in November last year the media

coverage of AI really

exploded with the release of chat GPT by

open AI so chat GPT is a type of AI

called a large language model or

generative

AI but it's not the only type of AI

there's other types of AI that are less

familiar to the general public like

machine learning computer vision natural

language uh processing and those AIS

mostly take in a single type of data and

we call this single model

AI so I'm going to give three Cutting

Edge examples of single model AI in

Healthcare the image behind me is an

image of a chest x-ray with a heart in

the center surrounded by the lungs the

ribs going across the shoulders at the

top AI has become really good at

distinguishing normal from abnormal we

call this

triaging and most X-Rays done in the

world are actually

normal so this software called chest

link by a company called oxy

pit it's it's a medical AI triage system

and it's the first system that's

received regulatory or C approval to be

used in a fully autonomous way um to

report on chest

x-rays so what oxy pit does is it looks

for 75 abnormalities on the chest x-ray

and if it doesn't find any of those

abnormalities it reports the X-ray as

normal without any human involvement if

it does find an abnormality it passes

the X-ray back to the human radiologist

to report this is an example of task

sharing between the AI and the human

radiologist the picture behind me is a

retina this is the tissue at the back of

your eye if you've ever been for an eye

test this is what the optician sees the

optician is looking for reversible

causes of blindness like macular

degeneration a group of researchers in

University College London developed an

AI model trained on 1.6 million pictures

of the retina this model is able to

diagnose eye disease and predict out

comes from eye conditions like macular

degeneration and that's very impressive

that it can do what most non-specialist

doctors find hard but it doesn't stop

there we think about a condition like

Parkinson's disease we don't think about

the back of the eye Parkinson's disease

affects your movement causes a Tremor

affects your walking the same AI model

can have a can look at the back of your

eye and predict Parkinson's disease

years before patients develop symptoms

so now not only is it uh can it see what

the human can see but it can see things

that the human can't

see however this model will never

diagnose Parkinson's disease and it

definitely will never give Compassionate

Care for Parkinson's

Disease AI like this must be used in

conjunction with highly trained healthc

Care

Professionals I'm going to move away

from computer vision towards large

language

models in December of last year Google

released a medical large language model

called Med

Pam they trained their generic large

language model called Pam to perform

medical question answering and this is

the first time ever that a computer or

an AI model has passed a US medical

licensing exam with a passing score of

67% and in only three months later Med

Pam 2 the next version got a score of

86% this is expert level on that

exam if you have a smartphone in your

pocket multimodal AI is available to you

right

now four weeks ago open AI released the

multimodal version of chat

GPT and this is an example I gave us

last week where I passed in a picture of

an ECG this is the electrical activity

of the heart a very common test that we

do in the hospital and I gave it a

little scenario 60-year-old male

presented with palpitations that's a

sensation of your heartbeating in your

chest

he could feel his heart skipping beats

no past medical history not currently on

medications the attached picture is is

ECG what is the next step for this

patient now very unhelpful for this

presentation chat GPT told me that it's

a machine learning model and not a

physician and it can't give me medical

advice so I asked her to help a friend

out and told us I'm doing a tedex talk

about multimodal Ai and please play

along and it did exactly

that now although the the ECG analysis

wasn't perfect it was very very close

and the follow-up advice that it gave

was

perfect but this gets even better when

multimodal large language models are

trained on medical tasks and the best

example of this is med Pam m M from

multimodal released by Google in July of

this year Med Pam M takes multiple

different types of input pictures of the

skin pictures of chest x-rays pictures

of pathology text from Radiology

images and performs multiple uh medical

tasks so it's not perfect but the

Radiology report that generated from Med

Pam M was compared to a human

radiologist report report and the

blinded assessors prefer the med pamm

report in 40% of

cases so the things we need to implement

multimodal AI safely are trust

explainability and randomized clinical

trials in relation to trust there was a

survey done in the United States and

over half of the

respondents would feel anxious

if they knew their healthcare worker was

relying on AI for their medical uh

treatment in the same survey

75% of the

respondents feared that doctors were

going to integrate AI too quickly before

understanding the risk to patients so we

have a lot of work to do to bridge this

Gap the second thing is

explainability explainable AI opens up

the black box to tell us why it made the

output so in our research what we're

interested in is why did the AI model

pick a particular blood pressure

medication for high blood pressure

should we just go along with what the

model says or do we want to know why it

got to that

conclusion if if the output from the

model agrees with our world viiew then

we might just go along with it and not

question it and that's a very risky area

in medicine called confirmation by

the third thing we need is randomized

clinical

trials AI models must be tested in the

same way that we test

medicines for for for medicines we use

randomized control trials this is the

peak of evidence in in in medicine one

group receiving the AI model another

group not receiving the AI model and

follow them up to see who does better so

what's the missing

piece where does the art of medicine fit

in as medical students and Junior

doctors were always taught to First Look

at the patient you never interpret a

result um an x-ray an ECG without

knowing the context of the patient we

often call this the eyeball

test and this this has been tested where

patients coming to emergency departments

nurses would try out them as red yellow

or green from just looking at the

patient and this was shown to be more

accurate than sophisticated models so in

the future I see a world where a picture

or a video of the patient is also fed

into the multimodal

model so now looking back at myself my

12-year-old self standing nervously on

that

stage I can see the parallels between

then and now

back then I was looking for data to

solve the mystery of who robbed the

fictional hotel but now we're looking

for data for a more profound

reason and that's to make Healthcare

more efficient personalized and

accessible imagine a world where remote

corners of low and middle inome

countries that have no access to

Specialized Care can gain insights from

these

models that's a world the medical Ai and

especially Multi multimodal Medical AI

can help us create so as we look to the

future we have to prioritize compassion

and understanding we have to build this

relationship between Ai and the humans

to allow the doctors more time to spend

with the patients to understand them and

give them a better chance at health and

happiness thank

you

[Music]