Inria
Inria.fr
Inria Channel
Séminaire : Modélisation et Calcul Scientifique
Statistical models of currents for measuring the variability of anatomical curves, surfaces and their evolution
A 14h, Entrée libre
- Date : 8/12/2011
- Lieu : Salle de conférence du bâtiment 16
- Intervenants : Stanley Durrleman, ICM (Institut du Cerveau et de la Moëlle épinière) au sein de l'hôpital de la Pitié Salpêtrière
- Organisateurs : Irène Vignon-Clementel
Our work introduces a systematic approach to analyze and understand the large variability of anatomical shapes observed in medical images, for instance to distinguish between normal and pathologic structures. We developed methodological, numeric and algorithmic tools to address this question in a way that is independent of the data and the targeted application.
This approach is built on the concept of ‘currents’, which enables shape comparison without the need of point correspondences, whether these shapes are given as point sets, curves, surfaces or volumes and independently of their topology. We proposed an efficient algorithmic framework to make this model usable for the automatic investigation of large data sets. This toolbox contains new numerical schemes based on regular lattices, whose speed of convergence, and hence use, does not depend on the data themselves. We also introduced an approximation method based on the search for adaptive basis, which limits the combinatorial explosion due to the massive use of large data sets in population studies.
Then, we derived a statistical framework for the analysis of the variability of anatomical structures inferred from samples drawn from a population. This method detects common anatomical features across the samples and describes how they vary within the population. For the first time, the joint modeling of both the geometrical variability (captured by smooth deformations) and the residual variability (which captures non-diffeomorphic variations like the density of fibers for instance) is introduced. This method yields not only quantitative statistical measures but also an interpretable representation of the detected features.
Eventually, this work is extended to the statistical analysis of shape evolution from longitudinal data (each subject is observed several times at different ages). The proposed method combines the morphological differences between individuals with the possible variations of the dynamics of the growth. The measure of possible developmental delays opens up unique perspectives for the characterization of pathological cases via the detection of unexpected growth of some organs.
The presentation will focus on the different concepts from a methodological and numerical perspective. The interest of the approach will be illustrated on open anatomical problems such as the characterization of the sulcal folding patterns at the surface of the cortex, the measure of the variability of the white matter fiber bundles and the detection of differences in endocranial growth rate in great apes.
Mots-clés : Modélisation et Calcul Scientifique Séminaire Centre de recherche Inria Paris - Rocquencourt
