[Hors-Série Google] IA & Santé : comment la technologie peut-elle sauver des vies ?

This hot series is offered by Google. And Cilia could help us save lives. Yes, artificial intelligence is good at improving the accuracy of diagnostics, making care more accessible and allowing us to accelerate medical research. The stake is large, especially in this month of October in Rose, dedicated to blood cancer sensitization, and it is essential to remember that the diagnosis of this disease is a reality for thousands of women. In France, nearly 60,000 new cases are reported each year, and in 2022, more than 2 million women were diagnosed worldwide. Better understand how Cilia can help us fight this disease and many others. This is the play of this new episode of our podcast series Google. Today I welcome Professor Anne Vincent Salomon, medical pathologist at the Curie Institute and Director of the Universitaire Hospital Hospital, Cancer des femmes and Joëlle Barral, Director of the Fundamental Research in Artificial Intelligence at Google DeepMind. Hello. Hello. Hello, Salomon. Anne Vincent Salomon, according to you, can we talk more about revolutions or rather evolutions when we talk about dia in the field of health? The two, of course, the word "revolution" means "clack" and "inspire", a real real change of practice, but in reality it is also an evolution, an evolution in the means that we are given to access medical information, to access tools that will help the doctors and therefore it is really the two. And you have to evolve the mentalities to accompany this evolution-evolution. And I turn to you, Joëlle, in the field of health, how can it help us to accelerate research? So, LIA helps us at many levels. First, when you do scientific research, you often go and look at everything that has been done before you. So you have all the literature that you have to explore and tools like Notebook LM, for example, you allow to do that at high speed. So that's a very concrete use. After that, LIA also allows us to set up hypotheses and to do a certain number of things, we say "in silicosis", that is to say, before doing experiments. And that's very precious because obviously the experimental validation is still necessary, but it costs a lot. And so it allows them both to the daily life to have, for example, something that will help them write code faster, in a very concrete way, but also to put into resonance things that they would not necessarily think about, because they are papers or scientific domains that are further away. And so LIA allows us to do these discoveries of habits that are very fortuitous, perhaps to accelerate this serendipity. So Anne, from your point of view as a practitioner, what are the most promising benefits of LIA in terms of strategy, treatment and prevention? Winning time in medicine, there are always repetitive tasks that you have to do like a pilot of aircraft, at the operating blocks, radiologists, pathologists. We really have checklists where you absolutely have to not forget anything. There, it is obvious that the students are there to help not lose time with these repetitive tasks and not forget anything. Then we will win in precision analysis. I'm going to tell you in the future what this evolution is doing. And in terms of prevention, is it going to be a tool as well? So in terms of prevention, it can help to spot better, for example, the patients who have secondary effects risks. There are different types of prevention, primary, secondary and tertiary prevention. At the hospital, it is really rather secondary and tertiary prevention that we practice, that is to say, for example, to avoid the secondary effects of drugs and treatments that we are going to give if we know all the characteristics of the patients and that thanks to LIA models, we can see the patients who are going to develop more pulmonary fibrosis at the radiotherapy or more hematological reactions to a given cancer anticancer. Well, we will have won in security and precision of secondary prevention. So according to the public health care in France in 2023, we estimated that there were more than 400,000 new cases of cancer in France. So according to you, Anne, in what extent will LIA be able to help us fight this disease? So fighting this disease directly, we would indeed have to progress already in understanding the causes. The causes of cancer are multiple. And so typically, we are facing a problem where we need to analyze large databases of data, retrospective, of patients with cancer and to be able to have the power to analyze to see the risk factors. There, we are really going to win in knowledge. But obviously there, to model, as Joel said, we need all the bibliographic data. We need to exploit our databases. And there, we're really going to be helped by these tools. I have the impression that we are clearly a gap between research and practice. And this is good because you have formed a collaboration. You have announced a partnership together. So what does this new partnership bring specifically, Anne? So clearly, we are in a phase of preparation, of the scientific research that we are going to lead together. So obviously, working for a company like Google and a hospital like the Curie Institute and the Cancer Institute of Women, is to potentialize the knowledge of the two partners. And of course, we are going to try to respond to very precise scientific questions. Google has a computer and IA knowledge that will certainly integrate data more easily. And we are going to bring the biological knowledge of these tumors and the technique itself, which is really a biological technique and manipulation of the samples. So Joel, for you, these private public partnerships, what are the challenges, the benefits, what does it mean to you? So for us, especially in quite complex areas like health, it is absolutely key to work with all the actors of an ecosystem. And so we are happy to have this partnership with the Curie Institute, which allows our researchers to really confront themselves with the reality of an excellence hospital that has always been in the early stages and really to this intersection between clinical research and care for patients. Another aspect that seems very important to me in private public collaborations is the fact that we have the chance to see how a certain number of countries evolve in their approach to health, along with a scientific revolution that is that of the IA that we have the chance to live today. And so I think we also have a role to play to allow everyone to see a little the good practices and how the one and the other try to respond to these great challenges. I have a question. Is it possible for you to have this power of strike, so already this multiplicity of knowledge, but then this force of generation of data that Google can bring, it can allow you to find treatments almost on a measure for patients, something that we could not do before? Obviously, that's what we want. The precision medicine is essential to individually solve the problem of people facing cancer. We now have the chance to be able to use immunotherapy with chemotherapy, which remains an essential source to treat patients with cancer. Now, how to bring chemotherapy? To kill as little body and vital functions as possible, but to destroy tumors? It's really the challenge of very targeted new therapies. And this is where the multidisciplinary analysis of these tumors will make us hope for breakthroughs. So we are waiting for this multiple level of analysis, facilitated by the tools that we will have in our hands, of the knowledge of Google researchers who will come to work with scientific and medical researchers of the institution. We really hope that it will be synergistic and very creative. Without a doubt. In the other episodes of our series, we were able to see how far France benefits from talent in the field of medicine. How does it apply to medical research? Can you see the same thing? Yes, indeed, I think that France has the chance to be a country in which there is a great mathematical tradition and the great models of language and the revolution of the generative link that we see today really rely on mathematics. So we find a lot of French people in the different research bodies and in the start-ups that are really in the very first years of this revolution. I think that France also has the chance to have two great doctors and a very high quality doctor. And so the growth of the two can also allow us to be in the very first years of medical research and advances that we can do, especially against women's cancer. So just Anne, talking about the discovery of advanced and creativity, can it allow us to imagine a new creative treatment that we would not have thought about? All our life is based on the proteins that live in the cells and that act in the cells as real workers. So, better understanding the structure of the proteins and understanding their role in the cells allows us to find the flaws of the tumor cells and imagine new treatments. So it may be very short, but in any case, it is really like accelerating the understanding of the living to better understand the flaws of cancer, it is essential. And I don't think we have really discussed this yet. But there is this very scientific aspect and the use of IA to make research progress. And then the use of IA to make time gain, while being a very high level help. So we are going to gain time and it will leave, I hope, for the doctors. More time for what is irreplaceable, that is to say empathy and interest for the sick person. And while increasing the scientific level of the quality of the care that we are going to deliver. I see that we are talking about daily practices. And concerning the medical imagery, in particular, IA is already an assistant for radiotherapists. Joel, how can this diagnostic aid be generalized in the future? So indeed, we had a lot of examples, these last 10 years, for diagnostic aid in a punctual way. That is to say for a type of given tumor or for a specific model of imagery. You talked about radiotherapy, we have a lot of examples in mammography or to read x-rays, etc. And what we see with IA generative is that it is no longer just about putting a diagnosis on an image, but about putting the image and text in resonance. And so today we have models that are able, from an image, to generate a first aid of a report that the radiologist would have brought to write, or to look at the whole of a patient's file, both the images, the clinical tests, the results of blood tests, etc. and try to have a global perspective on a patient. So that is obviously a step further than what we have been able to bring to do so far in the field of diagnostic aid. The pathologist, the job I do, the radiologist does it too. He integrates the clinical data and then the images he sees. And indeed, he helps with mammography, ecography, IRM, scanner and integrate all these data. It's always a real challenge, it's exciting, it's the heart of medical activity. And if it can be facilitated by tools, then once again, it will be for the benefit of time gained. And it also interprets the training of tomorrow's doctors who will, we hope, be able to. And therefore to have understanding, a mathematical understanding quite pushed, I think the tools there, it is certainly very precious, but also how to train yourself in a medical system, in the course of a medical course, medical studies. So precisely, Anne, sometimes there is a fear that there will be a replacement of human expertise and even the aspect of the profession of a doctor. On the other hand, what should be his place for doctors and healers? So it's true that every discovery of new technologies, there is this fantasy such or such a specialty will be replaced by new technology. I have an age, obviously, to advance to have lived several revolutions like that of technology. Well, this one is of an unequal range, obviously. But in fact, we always realize that instead of replacing, it increases the capacity of understanding of the living which still still keeps us from extremely vast mysteries and real challenges of understanding. But in any case, I have the confidence that we will not replace. We will increase, change in the way we work, but it will enrich, it will enrich our way of working. That, I have the real conviction. In fact, the diagnostic tools must be extremely ergonomic so that they are adopted. And finally, you always have to find a way to make them have an added value to the daily life of the doctor. Because finally, the big drama in France currently is that we are not quite many and therefore the workload is very, very significant. So there is also, in a very pragmatic way, LIA must facilitate the medical practice of the doctor. Anne, you mentioned the number of practitioners and some territories where access to care is perhaps less abundant. In France, on the patient side, how do you think that the use of LIA can transform their experience in these territories? So the use of LIA, by gaining time, by increasing the possibilities of synthesis, diagnostics and access to information for the doctors, we can hope that it will give more time to take care of more patients. It will not respond immediately to the medical desert. But we can imagine that tools based on LIA technologies facilitate the relay by advanced practices, for example, with nurses, nurses, who can be a real relay. So we can actually hope that it will help, it will not replace. And that, there must be solutions found for medical deserts with the physical presence of supply of care, of course. And there too, it affects an economic model in terms of territories, how to measure the French territory to access to care. It is a vast question that is largely out of my expertise, but which is a real solution. And I believe that what you say about medical deserts, LIA can easily reduce the waiting time. So this is not the only solution. Yes, it is a very interesting path that must be considered in the current reflection. So to finish, I would like to call on your imagination. If there was no constraint of feasibility, what would you like LIA to accomplish today that it does not? First of all, maybe just to ... I think it is important, LIA can do a lot today. It is extremely impressive and it still advances at an always so impressive speed. But it is not a magic baguette. And I think that does not answer your question, but I think it is important to remind them. But here, while posing this preambule, LIA will allow us to do giant bonds. And something that we did not talk about, which is ... you mentioned a little bit in the relationship with the doctor. But there is obviously an essential biological component in all diseases. There is a psychological component. And we do not necessarily always understand the intersection of the two. And why the psychological feeling of a patient is so important and also influences the course of a disease. And so I think that LIA will also bring us to better understand how the two work together. We also see malicious interactions with the conversational agents. Because we think they replace a human being, which they are completely unable to do. But coming to the personnalization that you are talking about, to bring to each one a help, in the end, all along his personalised, permanent illness. And where we manage to evolve, to evolve during the time, the benefit brought to the patient. I think that is something that I would like to continue to explore a lot. What I would like in the country in which we are, of course, is to tell me that these tools will be accessible to the largest number of public hospitals and doctors in the public system. It would be a dream that these tools, whose biases are corrected or are mastered, well understood. And that the training of students and doctors to use these tools can really be done. So I have a very practical dream. And this goes through means in hospitals, but obviously, hospitals have so many other urgent needs. In fact, you have to accompany progress and accompany this revolution of LIA, while maintaining this excellent system of French medicine and therefore in all its components. Then I find that precisely, that is what is exciting, to accompany this path. So in fact, to progress it, I think that what always makes me the most dreamy, is to continue to learn every day. That is what is absolutely exciting, and to know that we will be able to learn with tools that facilitate learning, but that make it even more of a high level, this enthusiasm. And maybe to get back to that, we also have the chance to be at the beginning of this revolution. And so there are some decisions that we are going to take now that will really have long-term consequences for French medicine, for the way students are going to be prepared in the coming decades, etc. And so I think that's what is also exciting and that makes us carry an important collective responsibility. And it joins together what we were saying about the challenges we want to tackle first, and how we are going to co-develop to really bring relevant solutions to the largest number. Thank you very much to both of you for your enlightening, it was exciting. Thank you very much. Thank you Estelle. Thank you. Thank you Anne. As you have been able to discover science and technology, you will find a line of creatives, a line on which the different public and private actors must advance with expertise by finding a just balance. This is the end of this episode, we will meet again soon for new encounters and new perspectives on the IA. Take care of yourselves.