Explainable and interactive large-scale content-based image retrieval : a new FNRS Research Fellow for Axelle Schyns
Her career :
Divided during all her secondary education between literary subjects and scientific subjects, it was finally the latter that prevailed when, in September 2018, Axelle Schyns decided to start studying engineering at the University of Liège.
Hesitating at first between computer science and biomedical science, Axelle took classes in both options, before deciding in her third year in favor of computer science, attracted by the courses she had taken. Following this decision, she left biology aside to take a few courses in electricity, a subject that seemed particularly complementary to computer science. However, this abandonment was only temporary since she was determined to reconnect with the biomedical field as soon as possible. Which she did by carrying out her master’s thesis relating to content-based image retrieval based on histopathology images (a field of medicine relating to the analysis of tissues in order to study the impact of diseases), under the supervision of Professor Pierre Geurts and Doctor Raphaël Marée.
After a busy year, shared between Liège and Ottawa, where she completed an Erasmus stay in the first term, Axelle graduated as an Engineer in Computer Science, with a focus on Intelligent Systems, with the greatest distinction in June 2023. She then continued her career by starting a doctoral thesis as an FNRS candidate with the aim of continuing and deepening the work started during her thesis.
Her research :
Axelle's research takes place in a context of increased digitalization, present in all types of fields but in particular in the medical field.
More and more images are generated by various technologies such as scanners, these technologies themselves are subject to increased development in recent years, all contributing to an overall increase in data from medical imaging.
In particular, the field of digital histopathology has led to obtaining millions of Whole Slide Images (WSI), i.e. images of gigantic sizes and of several gigabytes, representing sections of various tissues. These WSIs present enormous potential for medical and cancer research, which is why several pan-European initiatives (https://bigpicture.eu/) have been launched.
These initiatives aim to build databases for WSI in order to facilitate the use of artificial intelligence-based algorithms to analyze them. However, if the existence of such databases is an essential element for the creation of such algorithms, other factors currently limit their use: the lack of annotations/labels, bias due to the origin/type of WSI, the interpretability of the results obtained (crucial in medicine),…
Axelle aims to develop deep learning algorithms to overcome these limitations, based on the content-based image retrieval (CBIR) technique. This technique consists of analyzing the content of images to determine their similarity to a given image in order to return those that are most similar to it. Axelle will apply this technique at several levels and in different ways, testing several types of learning: supervised, unsupervised, self-supervised, and several methods: deep metric learning, multiple instance learning, etc. The algorithms designed and their codes will be integrated into the Cytomine platform to increase the number of analysis tools already available. Cytomine is a platform for storing and analyzing WSIs. It facilitates scientific collaboration by offering the possibility of collectively annotating the same WSI.
