The future of AI in medicine | Conor Judge | TEDxGalway

The last time I stood on this stage of the town hall theater was 26 years ago. He was a handsome 12 year old boy. He 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 child with a stutter or a speech impediment and I was desperately trying to remember and say my lines during that play I spent 70% of my I've spent 26 years gathering information to solve the mystery of who robbed the hotel and not much has changed.

I now work as a medical consultant in the hospital for half of my time and as a senior lecturer in applied clinical data analysis at 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 the cause of the disease is 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 about that information and communicating with the patient, so The information or data I collect takes many forms, patients, blood pressure, their medical history, blood test results, and this imbalance in how healthcare is delivered. 7030 is well known around the world and in many specialties technology has made it worse with the introduction of the electronic health record, for example, which was designed to collect billing information and not to make medical care more efficient, for example.

What this additional administrative workload that doctors have to do reduces the FaceTime that they have with patients, the FaceTime that we were fundamentally trained for and the FaceTime that patients want, so the idea that it is worth spreading and What I'm going to share tonight is a possible solution to this. We have all heard a lot about the risks of artificial intelligence AR, but I want to introduce you to a new perspective where the responsible use of medical AI could help solve some of these problems and I am going to introduce a new type of AI called multimodal. multimodal ai ai is AI that takes data in many different forms text images numbers when I work as a doctor in the hospital I talk to the patient I listen to the patient I listen to his chest with a stethoscope I am palpating his abdomen, I am looking at the results of his blood tests.

This is multimodal, multimodal human intelligence, which means many different types of data, so in November of last year the media coverage of AI really exploded with the launch of GPT chat by Open AI, so GPT chat . is a type of AI called large language model or generative AI, but it is not the only type of AI, there are other types of AI that are less familiar to the general public, such as machine learning, computer vision, language natural, processing and those AIS mostly take a single type of data and we call this a single AI model, so I'm going to give three cutting-edge examples of a single AI model 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 crossing the shoulders on top. AI has gotten really good at distinguishing the normal from the abnormal. We call this grading and most of the x-rays done in the world are actually normal so this software called Chest Link from a company called oxy pit is an AI medical grading system and it's the first system that has ever received regulatory approval or C to be used completely autonomously to report chest x-rays, then 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 involvement human.

If it finds an abnormality, it returns the x-ray to the human radiologist to report. This is an example of task sharing between the AI ​​and the human radiologist the image behind me is a retina this is the tissue in the back of your eye if you have ever been to an eye exam this is what the eye sees optician the optician is looking for reversible causes of blindness such as macular degeneration A group of researchers from University College London developed an artificial intelligence model trained on 1.6 million images of the retina. This model is capable of diagnosing eye diseases and predicting the outcomes of eye conditions such as macular degeneration, and it is very impressive that it can do what most do not do.

Medical specialists find it difficult, but it doesn't end there. We think of 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. It affects the way you walk. The AI ​​model itself can look like this. in the back of the eye and predicts Parkinson's disease years before patients develop symptoms, so now not only can it see what the human can see, but it can also see things that the human cannot see; however, this model will never diagnose Parkinson's disease. and will definitely never provide compassionate care for Parkinson's disease.

AI like this should be used in conjunction with highly trained healthcare professionals. I'm moving away from computer vision towards large language models. In December last year, Google launched a large medical language. model called Med Pam, they trained their generic big language model called Pam to answer medical questions and this is the first time that a computer or artificial intelligence model has passed a US medical licensing exam with a passing score of 67% and in just three months later, Med Pam 2, the next version scored 86%, this is the expert level in that exam, if you have a smartphone in your pocket, multimodal AI is available for you now same.

Four weeks ago, Open AI released the multimodal version of GPT chat and This is an example I gave us last week where I passed on an ECG image. This is the electrical activity of the heart, a very common test that we do in the hospital, and I presented a little scenario that a 60-year-old man presented. with palpitations, it is a sensation of your heart beating in your chest, you could feel your heart skipping beats, no medical history, not currently taking medications, the attached image is an ECG, what is the next step for this patient , now very useless for this presentation, the chat told me GPT. that he is a machine learning model and not a doctor and he can't give me medical advice so I asked him to help a friend and told her that I was giving a tedex talk on multimodal AI and to please play along and he did exactly that.

Now, although the ECG analysis wasn't perfect, it was very, very close and the follow-up advice he gave was perfect, but this gets even better when training large multimodal language models on medical tasks and the best example of This is

medicine

Pam. m M of multimodal launched by Google in July this year Med Pam M takes multiple different types of input skin images chest x-ray images pathology images radiology image text and performs multiple medical tasks, so it is not perfect, but The radiology report generated from Med Pam M was compared with the report of a human radiologist and the blind evaluators prefer the med pamm report in 40% of the cases, so what we need to implement the Multimodal AI safely is a reliable explainability and randomized clinical trials in relation to I am confident that a survey was conducted in the United States and more than half of the respondents would feel anxious if they knew that their healthcare worker depended on AI for your medical treatment.

In the same survey, 75% of respondents feared that doctors would integrate. AI is too fast before understanding the risk to patients, so we have a lot of work to do to close this gap. The second thing is explainability. Explainable AI opens the black box to tell us why it got the result in our research. What we are interested in is why the AI ​​model chose a particular blood pressure medication for high blood pressure, should we just accept what the model says or do we want to know why it came to that conclusion if the model's output agrees with our worldview then we could just accept it and not question it and that is a very risky area in

medicine

called confirmation.

The third thing we need is randomized clinical trials. AI models should be tested the same way we test medications. medicines we use randomized control trials this is the peak of evidence in medicine one group receives the AI ​​model another group does not receive the AI ​​model and we track to see who does it better so what is the missing piece ? Where does the art of medicine fit in? As medical students and young doctors were always taught to look at the patient first, you never interpret a result, an x-ray, an ECG without knowing the context of the patient, we often call it the eyeball test and this is has been tested where patients arriving in emergency departments, nurses would test them in red, yellow or green just by looking at the patient and this was proven to be more accurate than sophisticated models, so in the

future

I see a world where you can also see an image or video of the patient. fed into the multimodal model, so now, looking 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 stole the fictional hotel, but now we are looking for data for a deeper reason and that is to make healthcare more efficient, personalized and accessible.

Let's imagine a world where remote corners of low- and middle-income countries that don't have access to specialized care can get insights from these models. That's a world. Medical AI and especially multimodal medical AI can help us create, so as we look to the

future

, we must prioritize compassion and understanding. We have to build this relationship between AI and humans to allow doctors to spend more time with patients. understand them and give them a better chance at health and happiness thank you