Health / Personalised medicine

Octobre Rose: computer science and mathematics in the service of health

Date:
Changed on 07/10/2024
October Rose is a month dedicated to raising awareness and fighting breast cancer, and the fight against this disease is at the heart of health concerns. But it's not the only cancer taking its toll on women. The Monc project-team at the Inria Center at Bordeaux University is developing innovative mathematical modeling methods to optimize the analysis of microscopy slides, enabling the anticipation of responses to treatment and the risk of relapse in particular.
Octobre rose
© Freepik

Breast cancer: anticipating the response to treatment or the risk of relapse

In France, breast cancer remains a major health issue, representing the second leading cause of death in women after stroke. In 2023, Santé public France recorded almost 12,000 deaths linked to this pathology. Accurate diagnosis and treatment are therefore crucial to improving survival rates for many women.

The Monc project-team at the Inria Centre at the University of Bordeaux, which specializes in mathematical modeling for oncology, is working in particular on this type of disease. The team's scientists analyze histological slides from biopsies or surgical specimens. These slides, stained to reveal nuclei and cells, provide extremely high-resolution images (up to 100,000 pixels by 100,000). The colossal size of these images makes them difficult to analyze in their entirety, but thanks to the work of the Monc project team, it is now possible to extract essential information from them.

To process these images, the method consists of dividing them into tiles, i.e. subsets of smaller images, thus facilitating their management. Each slide can be sliced into around 10,000 tiles, enabling classification based on a representative sample. The team's scientists then use multi-instance deep learning, an artificial intelligence technique in which a computer learns from many small parts (in this case, the tiles) of a large image (the slide), in order to understand and make decisions on the whole without having to analyze everything at once. This process makes it possible to characterize the biological profile of samples and anticipate response to treatment or the risk of relapse.

Another specific objective is to study alternatives to commercial risk markers for certain types of breast cancer, in order to predict the probability of recurrence or response to cancer treatment. By developing methods to identify these markers directly from histological slides, scientists could potentially reduce the associated costs for public hospitals and make this information more accessible, greatly improving patient comfort.

Uterine cancer: overcoming diagnostic problems with the algorithms

Uterine cancer also represents a serious threat to women's health, being one of the leading causes of gynaecological mortality. This pathology is characterized by abnormal cell growth in the endometrium, the lining of the uterus, and requires special attention due to its significant impact on quality of life and available treatment options. In addition, some cases are difficult to diagnose, making it difficult to determine their degree of aggressiveness. To overcome this, biopsy slide analysis, once again, is at the heart of this research. However, the small number of patients necessitates a multi-center approach, involving slides from different hospitals and even different countries, each with different staining methods. In France, for example, saffron staining is common but expensive, making its use difficult to generalize internationally, despite the fact that it provides more precise details.

To harmonize data from different sources, the Monc project-team is working on algorithms capable of normalizing staining differences between slides. This makes it possible to apply robust multi-set learning methods despite the variations. 

Translational research

Understanding cancers is a complex issue that mobilizes research in many disciplines, far beyond medicine, such as biology, physics and the human and social sciences. Through its expertise in mathematical modelling, in close interaction with biological and clinical data, the Monc project-team conducts research on a variety of aspects ranging from biology to therapy guidance and treatment personalization.

The team draws on a combination of methodologies, including machine learning, differential equation modeling, image analysis and data assimilation. This diversity of approaches lies at the heart of translational research, with an impact that extends from mathematics to clinical application.

Equipe Monc
© Inria / Photo H. Raguet