Abstract
We address the problem of extracting meaningful, uncorrelated biological
modes of variation from
tangent space shape coordinates in 2D and 3D using non-Euclidean metrics. We
adapt the maximum autocorrelation factor analysis and the minimum noise
fraction transform to shape decomposition. Furthermore, we study metrics based
on repated annotations of a training set. We define a way of assessing
the correlation between landmarks contrary to landmark coordinates. Finally,
we apply the proposed methods to a 2D data set consisting of outlines of
lungs and a 3D/(4D) data set consisting of sets of
mandible surfaces. In the latter
case the end goal is to construct a model for growth prediction and simulation.
Original language | English |
---|---|
Title of host publication | Medical Image Computing and Computer-Assisted Intervention - MICCAI 2002, 5th Int. Conference, Tokyo, Japan |
Publisher | Springer |
Publication date | 2002 |
Publication status | Published - 2002 |