The team led by Sylvain Cussat-Blanc, University Professor in Computer Science at UT Capitole and winner of the 2024 edition of the Institut Universitaire de France (IUF), has been working for several years with Salvatore Valitutti, a researcher at the CRCT and a specialist in the immunological synapse, to develop a new algorithm that learns how to segment biomedical images.
This method, based on genetic programming and called Kartezio, is explained in the article ‘Evolutionary design of explainable algorithms for biomedical image segmentation’, published on 6 November 2023 in the multidisciplinary scientific journal ‘Nature Communications’.
This major advance in cancer research has just been awarded first prize in the Humies Awards at the Genetic and Evolutionary Computation Conference (GECCO) 2024.
 

Salvatore Valitutti's team has been working with members of IRIT on various research topics for almost 10 years. Over the last 4 years, an IRIT@CRCT team has even been set up and seven IT researchers have taken up permanent residence on the Oncopole premises. This cooperation enables them to better understand the needs of doctors and biologists in order to develop tools that can then be used.
 

The new technique developed by Sylvain Cussat-Blanc's team responds to a specific problem formulated by Salvatore Valitutti :

the difficulty of segmenting, in other words cutting out specific objects in an image, and more specifically images of immune system cells, cytotoxic T lymphocytes, which fight cancer cells. The new algorithm makes it possible to detect these cells and better understand how they interact with each other.
 

Based on an innovative genetic programming approach, Kartezio is capable of automatically generating biomedical image analysis pipelines from existing analysis functions.
This new method makes it possible to segment biomedical images with performances comparable to those obtained using deep learning techniques, but with a much smaller number of training images - only a dozen or so.

Another advantage of this technique is that the analysis pipelines produced are fully interpretable, a major asset for researchers and clinicians seeking to understand and justify the results obtained.
 

The Humies Awards recognise artificial evolution-based approaches that compete with humans in solving complex problems. Kartezio is proving to be more competitive than human operators in terms of both segmentation time and the accuracy of the segmentation performed.

Our approach opens up new perspectives for medical image processing, while developing computing techniques that push back the frontiers of the scientific state of the art.

This international recognition demonstrates not only the relevance and robustness of this work in the field of biomedical image analysis, but also the ability of the two institutes, IRIT and CRCT, to innovate by combining IT and medical expertise.