Algorithms multiply the power of imagery – Sciences et Avenir

Global health, the new challenge.
This first edition of the Rencontres du Grand Est will take place in Strasbourg on November 25, 2022. This event organized by Science and Future – Research is free, open to all; registration is required on the official website This registration opens the doors of the headquarters of the Grand Est region to attend these Meetings in “face-to-face”. It also gives access to the video that will allow you to attend this event live from the web, for those who do not have the possibility of being there on “D” day.

Global Health will be the opportunity to attend five conferences (“What is global health”, “Digital, robotics and AI in medicine and health”, “Harmful addictions for the hygiene of life”…) , three round tables (“Avoiding another pandemic. The effects of climate on health”, “Digital, AI, doctors and patients”, “Eating well, exercising and sleeping better”), two dialogues, and eight presentations by start-ups. In parallel, videos devoted to trees, planetary health, global health or even bats will be shown throughout the day. Among the speakers: Benjamin Roche (IRD), Bernard Nordlinger (Academy of Medicine), Irène Buvat (Institut Curie), Mathilde Pascal (Public Health France).

Santé Globale will be held at the headquarters of the Grand Est region, 1, place Adrien-Zeller, 67000 Strasbourg. November 25 at 9:30 a.m.

Automate the best possible diagnosis of cancers for the greatest number, from simple imaging, without biopsy or invasive examination. This is the incredible promise of radiomics, a discipline born in scientific publications only in 2012 – two studies that year, 912 in 2019 and more than 2200 in 2022. For just ten years, this imaging technique medicine trains algorithms to reveal, in simple images, not only the organization and architecture of tissues, but also their cellular and molecular composition. However, tumors are not just a mass of undifferentiated cancerous cells, but complex environments, each characteristic of which may prove relevant in choosing the most suitable treatment. The starting hypothesis of radiomics is thus particularly suited to cancer, the idea being to use artificial intelligence to extract genetic, protein, metabolic information from an MRI, a scanner or a PET scan , physiological and, of course, anatomical.

“There are a lot of proofs of concept, but no radiomics tool is yet in routine use”recalls Irène Buvat, director of the Laboratory of Translational Imaging in Oncology at the Curie Institute, in Paris, who will speak on this subject during the Grand Est Meetings on November 25 in Strasbourg (Bas-Rhin). “We still lack perspective to guarantee that the machines will give useful information for treatment, whatever the cancer “, adds the expert winner of the Pink Ribbon Avenir 2021 prize awarded to researchers advancing the fight against breast cancer.

Algorithms to determine the characteristics of tumors

Still, artificial intelligence can indeed be trained to reveal seemingly invisible information in digitized images. Thus, two algorithms working on anatomopathology slides, these samples of tumors taken during surgery or biopsy, have just obtained their CE marking, which paves the way for their marketing in Europe. Unlike radiomics which works on medical imaging live, taken from the patient, these two algorithms from the Franco-American start-up Owkin analyze tumor samples to determine their characteristics. In this case, the RlapsRisk BC program is designed to predict the likelihood that a person with early breast cancer will relapse after treatment, allowing oncologists to determine which high-risk patients may benefit from targeted therapies and which low-risk patients could avoid chemotherapy.

On this colorectal tumor sample, the algorithm of the start-up Owkin identifies the most genetically unstable cells (red dots).

On this colorectal tumor sample, the algorithm of the start-up Owkin identifies the most genetically unstable cells (red dots). (OWKIN)

The second, MSIntuit CRC, identifies a biomarker on a colorectal tumor sample revealing a defect in the cells’ ability to correct errors that occur when DNA is copied. Crucial information to administer the treatment with the best chance of success. To develop these two tools, Owkin collaborated with the Gustave-Roussy Institute (Villejuif) which provided thousands of these digitized anatomopathology slides as well as the corresponding patient data: response to treatments, genetic, clinical information, etc.

“With AI, a small hospital treating few cancers could offer diagnosis as good as what is done in the best centers

“This is the principle of supervised learning: we indicate to the program characteristics a priori non-visualcorresponding to the images it analyzes. In this way, he learns to interpret them beyond this simple visual aspect. “, specifies Irène Buvat. The difference between anatomopathological analysis on tumor samples and radiomics is that the first is not necessarily representative of the whole of the cancer since it only analyzes a sample. . “The advantage in both cases is not necessarily to make better than the best specialists, warns Irene Buvat. If they do as well, that would already make it possible to erase the inequalities that exist between different hospitals. With such a tool, a small hospital treating few cancers in the year could offer a diagnosis as good as what is done in the best cancer centers. “, and therefore a better chance of survival. In addition, the automation allowed by its programs could reduce the time taken for treatment.

“It is important to insist on the fact that AI is only a tool, that it will not replace doctors, contrary to what some colleagues sometimes like to say “, recalls Professor Bernard Nordlinger, who heads the “Artificial Intelligence and Health” working group at the Academy of Medicine and who will also be speaking in Strasbourg. “But practitioners will have to be trained in its use, from radiologists to general practitioners. The health of the future will be taken care of by augmented doctors in some way, but not automatic doctors “, he concludes.

Until then, it will be necessary to develop systems for evaluating the algorithms themselves to try to understand how they reach their conclusion. It is indeed the blind spot of AI, crucial in the field of care: if we know from what the machine learned and the results it gives, the how remains a mystery, often designated “black box effect” in the field. A point which will also be debated in Strasbourg on 25 November.

We want to thank the writer of this write-up for this remarkable material

Algorithms multiply the power of imagery – Sciences et Avenir

You can find our social media pages here and other related pages here