In cancer treatment, “artificial intelligence can be wrong” but “will progress”, says a researcher

How can artificial intelligence help medicine?
It can be very useful for analysis. The image is pixels: quantitative data that lends itself very well to exploitation by algorithms. It takes a lot of images and therefore a lot of patients to train the algorithm; we also integrate clinical data such as age, sex or the size of the lesion.

We pass all these data to the grinder, for what result?
An artificial intelligence algorithm could, tomorrow, help either the doctor to become more efficient or faster, or the patient to benefit from a more effective treatment. Adapted to everyone, depending on the characteristics of their cancer. Even though great progress has been made, there are still a lot of situations where, finally, we give the treatment, we evaluate its effectiveness and, if that does not work, we give a second line of treatment.

What can result in a loss of time, therefore a loss of opportunity?
If the initial treatment does not produce the expected result, it would have been better to go directly to chemotherapy. Except that today, there is immunotherapy, which works very well, but for 50 to 60% of patients. For the others, so 40%, we could have gone directly to something more effective. Except that, for that, we need what we call biomarkers, capable of telling us objectively whether the patient will respond or not. And these biomarkers, in immunotherapy, we don’t have them yet.

What are your lines of research?
The image we are working on, the pet-scanner, is very rich in information. Location and volume of the tumour, metabolic activity, elements of the patient’s immune system… Me, a radiologist, I measure the size of the cancer, I describe where any metastases are, but it’s a very visual analysis, which I can’t not output a lot of parameters. While the AI ​​algorithm, by taking the image at the pixel scale and analyzing it without a priori, gives a result much more relevant than what our eye is capable of. It is rather, today, an aid to the doctor. For example, for the automatic detection of a pulmonary nodule on a scanner, or the automatic classification of an anomaly on a mammogram: benign or malignant. On the other hand, artificial intelligence has great self-confidence, while it is capable of making mistakes. The doctor must therefore check the result. AI is going to progress, that’s the sense of the story, but it really offers a combination of computing power capabilities and doctor’s hindsight, which puts the information in context.

Can patients already benefit from it?
Tomorrow, we would like algorithms to bring us more than what today’s tools already give us. What we want is to predict the response to treatment. Predict survival. But we are still in the research stage. Changing the treatment by an AI, today, no. But tomorrow, yes. We can think that we will have these first tools in the course of the decade. They will arrive gradually in three, four years, but in clinical research. Before being able to be diffused in the routine, in a little ten years. AI and health is not just a fad.

This will be the doctor’s daily life in 2030?
After having succeeded, by then, in making research interdisciplinary. Make doctors, mathematicians and, of course, computer scientists work together. What we do at the 3IA institute (supported by the Côte d’Azur University, the 3IA Côte d’Azur institute brings together major players such as the CNRS, Inria, Inserm, Eurecom and many others, with the support of the CEA, the Nice University Hospital or many companies and local authorities, such as the Sophia Antipolis Agglomeration Community and the Nice Côte d’Azur Metropolis., editor’s note). The idea being to bring the algorithms developed by mathematicians to the patient’s bedside. These algorithms, doctors must have the ability to understand and criticize them. However, they do not do mathematics or algorithms in their studies. We have therefore set up a university degree to train them in understanding the mechanisms of AI. This course has met with success, with twenty-five enrolled in the first year.

We want to say thanks to the writer of this short article for this awesome material

In cancer treatment, “artificial intelligence can be wrong” but “will progress”, says a researcher

You can find our social media accounts as well as other pages that are related to them.