Procedural Generation of Anatomical Detail
Master Internship, IMAGINE research team.
Goal: Inventing new procedural modeling tools to turn anatomical knowledge into geometry generation programs.
Contact: François Faure
Current anatomical models such as ZygoteBody are impressive but they are coarse, static, and not customizable. We have recently addressed the animation and customization issues through the Anatomy Transfer paradigm, by fitting a rigged reference anatomy to a target shape, as shown in the image below, and in the video.
The level of detail is still coarse. For instance, only the main veins and arteries are represented, but the majority of the blood system, composed of thin to tiny vessels, is not available. This is also true for nerves, muscle internal structures such as fibers and their connections to the tendons, the multiple layers of the skin, connective tissues, etc. However, these mid-size entities are very important for advanced understanding of the anatomy as well as various simulations such as surgery planning or rehearsal.
The goal of this internship, which may be continued as a PhD thesis, is to tackle these challenges by generating the anatomy on-the-fly using novel procedural methods. The geometry would be generated while we zoom in or operate on a specific region of the body, and deleted when we move away.
The anatomical entities we want to generate have been highly studied, and their is a lot of knowledge available in the literature about their geometrical structure and their distribution in the human body. We tightly collaborate with the Anatomy Lab of the Grenoble University Hospital, who will provide us with the necessary help. A possible approach is to find inspiration in methods for the procedural generation of plants by combining qualitative knowledge of the geometry with statistical data about the distribution.
This work may be continued as a PhD thesis. The results may be used by the Anatoscope startup company.
- Master student in Computer Science or Applied Mathematics.
- Programming skills in C++ or python or java are mandatory.
- Background in mathematics (especially linear algebra, geometry, and statistics).
- Prior knowledge in the areas of computer vision, computer graphics and embedded programming are relevant.