2010 Summer Project Week Shape Analysis UNC
Key Investigators
- MIT: Tammy Riklin Raviv
- BWH: Sylvain Bouix
Project
Objective
Group-wise statistical analysis of brain structures is significant for clinical and neuroanatomy studies. Nevertheless, due to the inherent difficulty to properly represent the morphology of the structures of interest, the statistics is usually limited to volume and surface area measurements. Ideally, we would require an injective shape representation that is invariant rigid or even affine transformations and form a vector space. In the absence of shape representation that meets all of these requirements, it is difficult to define fundamental concepts such as the median or average shape.
We defined a novel measure for shape dissimilarity that is based on shape boundaries. This measure allows us to construct the 'median' of shapes of the same class that provides reacher morphological characteristics of the ensemble. Our goal in this project is to apply the proposed method to a population study of brain structures and test its discriminative power.
Approach, Plan
We use a variational framework based on level-sets to deform a representative (canonical) shape such that the sum of its distances from each of the shapes in the group is minimized. We will first test the approach on a set of synthetic shapes where the resulting median shape can be validated. The outcomes of this preliminary stage are expected to guide us in planning more advanced experiments.
