WT2: Extraction of Information from Images
Work Task Leader: INRIA
Main Contributors: UNIGE, UCL, STAR
The goal of this WT is to provide algorithms that efficiently extract organ shape, structure and motion from medical images. This task is coordinated and mainly carried out by INRIA. Other teams brought their own expertise (in the form of training, data and algorithms) to improve information extraction:
- UNIGE closely collaborate with INRIA to optimize various aspects involved in the segmentation: deformable model simulation, collision handling, levels of detail, shape priors, etc.
- STAR supply INRIA and UNIGE with a semantic-based data structure of the musculoskeletal system and high-level statistical descriptors for improved segmentation (topological and example-based constraints).
- UCL define the relevant information to be extracted/individualized from images, support models' validation, and adapt the acquisition protocols (WT1) according to the feedback from INRIA.
Major Research achievements
Full lower limb musculoskeletal modeling with associated anatomical taxonomy.
Design of novel segmentation techniques based on appearance (multimodal intensity profile clustering) and shape (multi-resolution statistical shape models) priors.
Exploration of GPU-assisted approaches to increase speed and interactivity of segmentation (in collaboration with CRS4 in WT7).
Study of collaborative, network and scalable tools for collaborative segmentation.