Extraction of 3d anatomical point landmarks based on invariance principles

摘要

We describe 3D operators for extracting anatomical landmarks which are based on only first-order partial derivatives of an image. To improve the predictability of the extraction results we analyze certain properties of the operators. First, we provide a statistical interpretation in terms of the Cramér Rao bound representing the minimal localization uncertainty. Second, we show that the operators can be derived on the basis of invariance principles. It turns out that the operators form a complete set of principal invariants. Third, we analyze the detection performance using a certain type of performance visualization and a scalar performance measure. Experimental results are presented for 3D tomographic images of the human brain.