Supervisor: Dr. Vincent Detours
Project. Histology started with the invention of the microscope in the XVIIth century. It yielded fundamental discoveries such as the cellular basis of life. Yet, while microscope technology has massively improved since then, the form of histological structures description has barely evolved across the centuries: it rests on natural language statements that are qualitative and somewhat subjective. As a result, many differential diagnostics tasks in histo-pathology are notoriously unreliable. In addition, researchers lack a principled quantitative and objective framework to properly investigate histological images.
Our group has developed an artificial intelligence-based pipeline that sets histology of a firm quantitative ground. Given a large collection of whole slide images (WSI) it generates an atlas of the morphologies they contain and the quantifications of each morphology in each WSI. Importantly, this pipeline is unsupervised and free of human biases. We have already produced prototype atlases for several human organs in health and disease.
The successful applicant will:
- Introduce the concept of magnification/scale in the morphological analysis underlying the atlases.
- Generate automatic annotations of morphologies from vision/language LLM, and from datasets orthogonal to slide collections (e.g. clinical data, genomics, transcriptomics,…).
- Bringing the current pipeline to production level
- Bringing atlases reliability, features and aesthetics to levels suitable for distribution to the scientific community at large.
- Generate atlases from collaborators datasets.
A prototype histology atlas for the thyroid gland can be browsed here (not suitable for phones and tablets yet!):
https://www-hpda.ulb.ac.be/iribhm/ai/atlases/demo-atlas/Thyroid/graphics/sankey.html
Histological categories are organized by increasing granularity from left to right.
Keyword. Artificial intelligence, LLM, foundation models, digital pathology, histopathology, whole slide images
Skills. The applicant has a master computer sciences, mathematics, physics or bioinformatics and has
- Outstanding coding skills
- A working knowledge of web development
- A working knowledge of AI tools, including proven fluency with pytorch
- Is eager to learn about biomedicine and follow medical school training in histology
- Is able to communicate effectively in an interdisciplinary environment
Is able to multitask between several concurrent projects